(19)
(11)EP 3 882 095 A1

(12)EUROPEAN PATENT APPLICATION
published in accordance with Art. 153(4) EPC

(43)Date of publication:
22.09.2021 Bulletin 2021/38

(21)Application number: 20840576.1

(22)Date of filing:  17.07.2020
(51)International Patent Classification (IPC): 
B60W 30/08(2012.01)
G08G 1/16(2006.01)
B60W 30/14(2006.01)
(52)Cooperative Patent Classification (CPC):
B60W 40/105; B60W 30/14; B60W 30/08; G08G 1/16
(86)International application number:
PCT/CN2020/102644
(87)International publication number:
WO 2021/008605 (21.01.2021 Gazette  2021/03)
(84)Designated Contracting States:
AL AT BE BG CH CY CZ DE DK EE ES FI FR GB GR HR HU IE IS IT LI LT LU LV MC MK MT NL NO PL PT RO RS SE SI SK SM TR
Designated Extension States:
BA ME
Designated Validation States:
KH MA MD TN

(30)Priority: 17.07.2019 CN 201910646083

(71)Applicant: HUAWEI TECHNOLOGIES CO., LTD.
Shenzhen, Guangdong 518129 (CN)

(72)Inventors:
  • DU, Mingbo
    Shenzhen, Guangdong 518129 (CN)
  • TAO, Yongxiang
    Shenzhen, Guangdong 518129 (CN)

(74)Representative: Epping - Hermann - Fischer 
Patentanwaltsgesellschaft mbH Schloßschmidstraße 5
80639 München
80639 München (DE)

 
Remarks:
The applicant has filed a text with which it is intended to bring the translation into conformity with the application as filed (Art. 14(2) EPC).
 


(54)METHOD AND DEVICE FOR DETERMINING VEHICLE SPEED


(57) This application discloses a method and an apparatus for determining a vehicle speed. During determining of a traveling speed of a vehicle, a probability distribution of intentions may be computed based on observation information of a surrounding object. Then, a probability redistribution of the different intentions is computed based on travel times required for the vehicle to travel from a current position to risk areas corresponding to the different intentions. Further, motion status variations of the surrounding object with the different intentions are predicted based on the travel times required for the vehicle to travel to the risk areas corresponding to the different intentions. Finally, the travelling speed of the vehicle is determined based on the probability redistribution of the different intentions, the motion status variations of the surrounding object with the different intentions, and motion status variations of the vehicle under different travelling speed control actions. In this way, a high risk, between the surrounding object and the vehicle, that is less likely to occur is not ignored. Therefore, a determined travelling speed is more appropriate for a current driving environment, and a potential safety risk during travelling of the vehicle is reduced.




Description


[0001] This application claims priority to Chinese Patent Application No. 201910646083.4, filed with the China National Intellectual Property Administration on July 17, 2019 and entitled "METHOD AND APPARATUS FOR DETERMINING VEHICLE SPEED", which is incorporated herein by reference in its entirety.

TECHNICAL FIELD



[0002] This application relates to the field of vehicle technologies, and in particular, to a method and an apparatus for determining a vehicle speed.

BACKGROUND



[0003] In a self-driving technology or the like, to avoid a collision between a vehicle and an object such as a surrounding pedestrian, a vehicle speed needs to be determined based on a motion status of the surrounding object. As the motion status of the surrounding object is affected by a subjective intention of the surrounding object, the vehicle may predict a possible intention of the surrounding object, to determine a vehicle speed based on a motion status of the surrounding object corresponding to a possible intention that can be learned through prediction. However, currently, during prediction of the intention of the surrounding object, impact caused by a risk of a collision between the surrounding object and the vehicle is usually ignored. Consequently, a determined vehicle speed is not appropriate enough, and there may be a safety risk when the vehicle travels.

SUMMARY



[0004] Embodiments of this application provide a method and an apparatus for determining a vehicle speed, so that a vehicle can determine a travelling speed of the vehicle based on a probability redistribution of intentions of a surrounding object, motion status variations of the surrounding object with different intentions, and motion status variations of the vehicle under different travelling speed control actions. In this way, a high risk, between the surrounding object and the vehicle, that is less likely to occur is not ignored. Therefore, a determined travelling speed is more appropriate, and a potential safety risk during travelling of the vehicle is reduced.

[0005] According to a first aspect, an embodiment of this application provides a method for determining a vehicle speed. The method may specifically include: first obtaining observation information of a surrounding object of a vehicle by observing the surrounding object of the vehicle; computing, based on the observation information of the surrounding object, a probability distribution of different intentions of the surrounding object; performing redistribution computation on the probability distribution based on travel times required for the vehicle to travel from a current position of the vehicle to risk areas corresponding to the different intentions, to obtain a probability redistribution of the different intentions, where the risk areas corresponding to the different intentions are areas, in a lane in which the vehicle travels, located proximate to the surrounding object with the different intentions; predicting motion status variations of the surrounding object with the different intentions based on the travel times required for the vehicle to travel to the risk areas corresponding to the different intentions; and finally determining a travelling speed of the vehicle based on the probability redistribution of the different intentions, the motion status variations of the surrounding object with the different intentions, and motion status variations of the vehicle under different travelling speed control actions.

[0006] It can be learned that, according to the method provided in this embodiment of this application, in a travelling process of the vehicle, for the plurality of possible intentions of the surrounding object of the vehicle, the probability distribution of the intentions may be computed based on the observation information of the surrounding object. Then, the probability redistribution of the different intentions is computed based on the travel times required for the vehicle to travel from the current position to the risk areas corresponding to the different intentions. Further, the motion status variations of the surrounding object with the different intentions are predicted based on the travel times required for the vehicle to travel to the risk areas corresponding to the different intentions. Finally, the travelling speed of the vehicle is determined based on the probability redistribution of the different intentions, the motion status variations of the surrounding object with the different intentions, and the motion status variations of the vehicle under different travelling speed control actions. In this way, during determining of the travelling speed of the vehicle, each possible intention of the surrounding object is considered, and a risk degree of a collision, between the surrounding object and the vehicle, that is corresponding to each intention and that is under control of each acceleration of the vehicle is further considered. In this way, a high risk, between the surrounding object and the vehicle, that is less likely to occur is not ignored. Therefore, a determined travelling speed is more appropriate for a current driving environment, and a potential safety risk during travelling of the vehicle is reduced.

[0007] With reference to a possible implementation of the first aspect, the computing, based on the observation information of the surrounding object, a probability distribution of different intentions of the surrounding object may include the following operations during specific implementation: establishing, based on the observation information of the surrounding object by establishing a coordinate system related to the lane in which the vehicle travels, a relative position relationship between the surrounding object and the lane and a relative motion relationship between the surrounding object and the lane; and computing the probability distribution of the different intentions of the surrounding object based on the relative position relationship between the surrounding object and the lane and the relative motion relationship between the surrounding object and the lane. In this way, the probability distribution of the different intentions of the surrounding object can be computed more conveniently and accurately through conversion of coordinate systems. This provides accurate data for subsequent determining of an appropriate vehicle speed.

[0008] With reference to another possible implementation of the first aspect, the method may further include: obtaining observation information of the vehicle; establishing, based on the observation information of the vehicle and the observation information of the surrounding object by establishing a coordinate system related to the lane in which the vehicle travels, a relative position relationship between the vehicle and the lane and a relative motion status between the vehicle and the lane, and a relative position relationship between the surrounding object and the lane and a relative motion status between the surrounding object and the lane; determining, based on the relative position relationship between the surrounding object and the lane and the relative motion status between the surrounding object and the lane, the risk areas corresponding to the different intentions; and computing, based on the relative position relationship between the vehicle and the lane and the relative motion status between the vehicle and the lane, and the risk areas corresponding to the different intentions, the travel times required for the vehicle to travel from the current position of the vehicle to the risk areas corresponding to the different intentions. In this way, during determining of the travelling speed of the vehicle, each possible intention of the surrounding object is considered, and a risk degree of a collision, between the surrounding object and the vehicle, that is corresponding to each intention and that is under control of each acceleration of the vehicle is further considered. In this way, a high risk, between the surrounding object and the vehicle, that is less likely to occur is not ignored. Therefore, a determined travelling speed is more appropriate for a current driving environment, and a potential safety risk during travelling of the vehicle is reduced. With reference to still another possible implementation of the first aspect, the performing redistribution computation on the probability distribution based on travel times required for the vehicle to travel to risk areas corresponding to the different intentions, to obtain a probability redistribution of the different intentions may include the following operations during specific implementation: performing particle processing on the probability distribution, where quantities of particles corresponding to the different intentions are used to represent the probability distribution of the different intentions; and adjusting, based on the travel times that are obtained through computation and that are required for the vehicle to travel to the risk areas corresponding to the different intentions, weights of the particles corresponding to the different intentions, to obtain the probability redistribution of the different intentions. In this case, to cover more surrounding objects and computes occurrence probabilities of all possible intentions of each surrounding object, a concept of a particle may be introduced. Through particle processing and computation, a risk degree of each intention can be determined based on a travel time required for the vehicle to travel to a risk area corresponding to the intention. In other words, a probability redistribution of the different intentions is obtained. This provides an indispensable data foundation for subsequent accurate determining of a speed of the vehicle and improvement of safety and reliability of a vehicle that uses an intelligent driving technology such as self-driving.

[0009] With reference to still yet another possible implementation of the first aspect, the predicting motion status variations of the surrounding object with the different intentions based on the travel times required for the vehicle to travel to the risk areas corresponding to the different intentions may include the following operations during specific implementation: determining, based on the travel times required for the vehicle to travel to the risk areas corresponding to the different intentions, probabilities that the surrounding object with the different intentions changes the current intention; and predicting the motion status variations of the surrounding object with the different intentions based on the probabilities that the surrounding object with the different intentions changes the current intention and a random probability.

[0010] With reference to a further possible implementation of the first aspect, the determining a travelling speed of the vehicle based on the probability redistribution of the different action intentions, the motion status variations of the surrounding object with the different intentions, and motion status variations of the vehicle under different travelling speed control actions may include the following operations during specific implementation: estimating travelling effects of the vehicle that bring about under the different travelling speed control actions based on the probability redistribution of the different intentions, the motion status variations of the surrounding object with the different intentions, and the motion status variations of the vehicle under the different travelling speed control actions; selecting a target travelling speed control action from the different travelling speed control actions based on the travelling effects of the vehicle that bring about under the different travelling speed control actions; and determining the travelling speed of the vehicle based on the target travelling speed control action. In this way, during determining of the vehicle speed, each possible intention of the surrounding object is considered, and a risk degree of a collision, between the surrounding object and the vehicle, that is corresponding to each intention and that is under control of each acceleration of the vehicle is further considered. In this way, a high risk, between the surrounding object and the vehicle, that is less likely to occur is not ignored. Therefore, a determined travelling speed is more appropriate for a current driving environment, and a potential safety risk during travelling of the vehicle is reduced.

[0011] According to a second aspect, an embodiment of this application further provides an apparatus for determining a vehicle speed. The apparatus includes a first obtaining unit, a first computation unit, a second computation unit, a prediction unit, and a first determining unit. The first obtaining unit is configured to obtain observation information of a surrounding object of a vehicle. The first computation unit is configured to compute, based on the observation information of the surrounding object, a probability distribution of different intentions of the surrounding object. The second computation unit is configured to perform redistribution computation on the probability distribution based on travel times required for the vehicle to travel from a current position of the vehicle to risk areas corresponding to the different intentions, to obtain a probability redistribution of the different intentions. The risk areas corresponding to the different intentions are areas, in a lane in which the vehicle travels, located proximate to the surrounding object with the different intentions. The prediction unit is configured to predict motion status variations of the surrounding object with the different intentions based on the travel times required for the vehicle to travel to the risk areas corresponding to the different intentions. The first determining unit is configured to determine a travelling speed of the vehicle based on the probability redistribution of the different intentions, the motion status variations of the surrounding object with the different intentions, and motion status variations of the vehicle under different travelling speed control actions.

[0012] With reference to a possible implementation of the second aspect, the first computation unit may include an establishment subunit and a computation subunit. The establishment subunit is configured to establish, based on the observation information of the surrounding object by establishing a coordinate system related to the lane in which the vehicle travels, a relative position relationship between the surrounding object and the lane and a relative motion relationship between the surrounding object and the lane. The computation subunit is configured to compute the probability distribution of the different intentions of the surrounding object based on the relative position relationship between the surrounding object and the lane and the relative motion relationship between the surrounding object and the lane.

[0013] With reference to another possible implementation of the second aspect, the apparatus may further include a second obtaining unit, an establishment unit, a second determining unit, and a third computation unit. The second obtaining unit is configured to obtain observation information of the vehicle. The establishment unit is configured to establish, based on the observation information of the vehicle and the observation information of the surrounding object by establishing a coordinate system related to the lane in which the vehicle travels, a relative position relationship between the vehicle and the lane and a relative motion status between the vehicle and the lane, and a relative position relationship between the surrounding object and the lane and a relative motion status between the surrounding object and the lane. The second determining unit is configured to determine, based on the relative position relationship between the surrounding object and the lane and the relative motion status between the surrounding object and the lane, the risk areas corresponding to the different intentions. The third computation unit is configured to compute, based on the relative position relationship between the vehicle and the lane and the relative motion status between the vehicle and the lane, and the risk areas corresponding to the different intentions, the travel times required for the vehicle to travel from the current position of the vehicle to the risk areas corresponding to the different intentions. With reference to still another possible implementation of the second aspect, the second computation unit may include a processing subunit and an adjustment subunit. The processing subunit is configured to perform particle processing on the probability distribution, where quantities of particles corresponding to the different intentions are used to represent the probability distribution of the different intentions. The adjustment subunit is configured to adjust, based on the travel times that are obtained through computation and that are required for the vehicle to travel to the risk areas corresponding to the different intentions, weights of the particles corresponding to the different intentions, to obtain the probability redistribution of the different intentions.

[0014] With reference to still yet another possible implementation of the second aspect, the prediction unit may include a first determining subunit and a prediction subunit. The first determining subunit is configured to determine, based on the travel times required for the vehicle to travel to the risk areas corresponding to the different intentions, probabilities that the surrounding object with the different intentions changes the current intention. The prediction subunit is configured to predict the motion status variations of the surrounding object with the different intentions based on the probabilities that the surrounding object with the different intentions changes the current intention and a random probability.

[0015] With reference to a further possible implementation of the second aspect, the first determining unit may include an estimation subunit, a selection subunit, and a second determining subunit. The estimation subunit is configured to estimate travelling effects of the vehicle that bring about under the different travelling speed control actions based on the probability redistribution of the different intentions, the motion status variations of the surrounding object with the different intentions, and the motion status variations of the vehicle under the different travelling speed control actions. The selection subunit is configured to select a target travelling speed control action from the different travelling speed control actions based on the travelling effects of the vehicle that bring about under the different travelling speed control actions. The second determining subunit is configured to determine the travelling speed of the vehicle based on the target travelling speed control action.

[0016] It may be understood that the apparatus provided in the second aspect corresponds to the method provided in the first aspect. Therefore, for implementations of the second aspect and technical effects that can be achieved by the implementations of the second aspect, refer to related descriptions of the implementations of the first aspect.

[0017] According to a third aspect, an embodiment of this application further provides a vehicle. The vehicle includes a sensor, a processor, and a vehicle speed controller. The sensor is configured to: obtain observation information of a surrounding object of the vehicle, and send the observation information to the processor. The processor is configured to: determine a travelling speed of the vehicle according to the method in any implementation of the first aspect, and send the travelling speed to the vehicle speed controller. The speed controller is configured to control the vehicle to travel at the determined travelling speed of the vehicle.

[0018] According to a fourth aspect, an embodiment of this application further provides a vehicle. The vehicle includes a processor and a memory. The memory stores an instruction, and when the processor executes the instruction, the vehicle performs the method according to any one of the implementations of the first aspect.

[0019] According to a fifth aspect, an embodiment of this application further provides a computer program product. When the computer program product runs on a computer, the computer is enabled to perform the method according to any one of the implementations of the first aspect.

[0020] According to a sixth aspect, an embodiment of this application further provides a computer-readable storage medium. The computer-readable storage medium stores an instruction. When the instruction is run on a computer or a processor, the computer or the processor is enabled to perform the method according to any one of the implementations of the first aspect.

BRIEF DESCRIPTION OF DRAWINGS



[0021] To describe technical solutions in embodiments of this application more clearly, the following briefly describes the accompanying drawings for describing the embodiments. Certainly, the accompanying drawings in the following descriptions show merely some embodiments described in this application, and a person of ordinary skill in the art may derive other drawings from these accompanying drawings.

FIG. 1 is a schematic diagram of a road traffic scenario related to an application scenario according to an embodiment of this application;

FIG. 2 is a schematic diagram of a hardware architecture of a vehicle that uses a self-driving technology or the like according to an embodiment of this application;

FIG. 3 is a schematic architectural diagram of a vehicle system that uses a self-driving technology or the like according to an embodiment of this application;

FIG. 4 is a schematic structural diagram of a vehicle that uses a self-driving technology or the like according to an embodiment of this application;

FIG. 5 is a schematic flowchart of a method for determining a vehicle speed according to an embodiment of this application;

FIG. 6 is a schematic diagram of intentions of a pedestrian according to an embodiment of this application;

FIG. 7 is a schematic diagram of a vehicle-surrounding object-lane model according to an embodiment of this application;

FIG. 8 is a schematic diagram of particlezation representation according to an embodiment of this application;

FIG. 9 is a schematic diagram of determining a risk area and a travel time according to an embodiment of this application;

FIG. 10 is a schematic diagram of an example of a surrounding object interactive motion model according to an embodiment of this application;

FIG. 11 is a schematic structural diagram of an apparatus for determining a vehicle speed according to an embodiment of this application;

FIG. 12 is a schematic structural diagram of a vehicle according to an embodiment of this application; and

FIG. 13 is a schematic structural diagram of another vehicle according to an embodiment of this application.


DESCRIPTION OF EMBODIMENTS



[0022] When a vehicle travels in a lane, a surrounding object such as a surrounding pedestrian or an animal needs to be considered in determining a vehicle speed, to avoid a traffic accident such as a collision with the surrounding object of the vehicle, and ensure safety of the vehicle and the surrounding object of the vehicle.

[0023] Currently, to better avoid the surrounding object such as a pedestrian, during determining of a travelling speed of a vehicle, considering that a motion status of the surrounding object is mainly affected by a subjective intention of the surrounding object, the vehicle may predict a target intention of the surrounding object by using behavior characteristics of the surrounding object. For example, the vehicle determines occurrence probabilities of various intentions of the surrounding object based on the behavior characteristics of the surrounding object, and sets a probability threshold, to select an intention with a relatively high occurrence probability as the target intention. Then, the vehicle determines the vehicle speed based on a motion status of the surrounding object with the target intention. However, in addition to the target intention, the surrounding object may also have another intention, and there may be a relatively large risk of a collision between the surrounding object and the vehicle with the another intention. Therefore, if only the target intention with a high occurrence probability is considered and the vehicle speed is determined based on the single target intention that is obtained through prediction, some other intentions that have a low occurrence probability but are highly likely to cause collision risks are ignored. In this case, once the surrounding object actually moves with the ignored intention with a high collision risk, and the vehicle travels at the travelling speed determined based on the target intention, it is very likely that the vehicle collides with the surrounding object. Consequently, a safety risk is caused.

[0024] For example, referring to a schematic diagram of a road traffic scenario shown in FIG. 1, a vehicle 101 drives by using a self-driving technology, and surrounding objects of the vehicle 101 in the scenario include a pedestrian 102 and a pedestrian 103. It is assumed that the vehicle 101 predicts respective target intentions of the pedestrian 102 and the pedestrian 103 by using behavior characteristics of the pedestrian 102 and behavior characteristics of the pedestrian 103. The prediction is as follows: The target intention of the pedestrian 102 is fast crossing a lane or diagonally crossing a lane in a driving direction, and the target intention of the pedestrian 103 is stop. In this case, a determined vehicle speed of the vehicle 101 is 60 km/h. However, otherpossible intentions of the pedestrian 102 and the pedestrian 103 that may cause collision with the vehicle 101 are not considered when the vehicle 101 predicts the intentions. Therefore, the target intention, obtained through prediction, of the pedestrian 103 that is closer to the vehicle 101 does not include an intention of fast crossing the lane or an intention of diagonally crossing the lane in the driving direction that is hardly likely to occur. In this case, the determined vehicle speed is relatively large. However, the pedestrian 103 may fast cross the lane or diagonally cross the lane in the driving direction. Because the target intention of the pedestrian 103 is predicted inaccurately, the vehicle 101 is highly likely to ignore an intention with a relatively high collision risk. In this case, when the vehicle 101 travels at the relatively high speed of 60 km/h, the vehicle 101 is highly likely to hit the pedestrian 103, causing a traffic accident related to the vehicle 101 and the pedestrian 103. Based on this, to resolve a problem that a determined vehicle speed is inappropriate because an intention of a surrounding object is not accurately or comprehensively predicted, in the embodiments of this application, a method for determining an appropriate vehicle speed is provided. The vehicle and a surrounding object of the vehicle can be safe provided that the vehicle travels at a determined appropriate speed. A specific process of determining the vehicle speed may include: computing, based on observation information of the surrounding object, a probability distribution of all intentions, and computing a probability redistribution of the different intentions based on travel times required for the vehicle to travel from a current position to risk areas corresponding to the different intentions; then predicting, based on the travel times required for the vehicle to travel to the risk areas corresponding to the different intentions, motion status variations of the surrounding object with the different intentions; and finally, determining a travelling speed of the vehicle based on the probability redistribution of the different intentions, the motion status variations of the surrounding object with the different intentions, and motion status variations of the vehicle under different accelerations. In this way, during determining of the vehicle speed, each possible intention of the surrounding object is considered, and a risk degree of a collision, that is corresponding to each intention, between the surrounding object and the vehicle that travels by using different accelerations are further considered. In this way, a high risk, between the surrounding object and the vehicle, that is less likely to occur is not ignored. Therefore, a determined travelling speed is more appropriate for a current driving environment, and a potential safety risk during travelling of the vehicle is reduced.

[0025] The scenario shown in FIG. 1 is still used as an example. Assuming that the vehicle 101 determines the vehicle speed by using the method provided in the embodiments of this application, a specific determining process may include the following: First, probability distributions b102 and b103 of seven intentions (including: stepping forward along a sidewalk, stepping backward along a sidewalk, straight crossing a lane, diagonally crossing a lane in a driving direction, diagonally crossing a lane in a direction opposite to a driving direction, stepping away from a lane, and stop) of the pedestrian 102 and the pedestrian 103 are predicted respectively based on observation information of the pedestrian 102 and observation information of the pedestrian 103. b102 includes a probability P102i (i=1, 2, ..., 7) that the pedestrian 102 has each intention, and b103 includes a probability P103i (i=1, 2, ..., 7) that the pedestrian 103 has each intention. Then, for each intention, an area, in the lane in which the vehicle 101 travels, located proximate to the pedestrian 102 with the intention is determined and recorded as a risk area corresponding to the pedestrian 102 with the intention; and an area, in the lane in which the vehicle 101 travels, located proximate to the pedestrian 103 with the intention is determined and recorded as a risk area corresponding to the pedestrian 103 with the intention. Then a travel time T102i required for the vehicle 101 to travel from a current position to the risk area corresponding to the pedestrian 102 is computed; and a travel time T103i required for the vehicle 101 to travel from the current position to the risk area corresponding to the pedestrian 103 is computed. After that, redistribution computation may be performed on the probability distribution b102 based on the travel time T102i, to obtain a probability redistribution b'102 corresponding to the intentions; and redistribution computation is performed on the probability distribution b103 based on the travel time T103i, to obtain a probability redistribution b'103 corresponding to the intentions. In addition, motion status variations of the pedestrian 102 with the different intentions and motion status variations of the pedestrian 103 with the different intentions may be further predicted respectively based on the travel time T102i and the travel time T103i. Finally, the vehicle 101 may determine a travelling speed of the vehicle as 30 km/h based on the probability redistribution b'102 and the probability redistribution b'103; the motion status variations of the pedestrian 102 with the different intentions and the motion status variations of the pedestrian 103 with the different intentions; and motion status variations of the vehicle under different accelerations. In this way, during prediction of the intentions of the pedestrian 102 and the pedestrian 103, various possible intentions are considered, and a relatively appropriate vehicle speed is jointly determined with reference to situations that may occur when the vehicle uses different coping strategies under each intention. Therefore, the vehicle 101 travels in the lane at a speed of 30 km/h, which is relatively safe. This effectively avoids a safety risk that may exist during travelling of the vehicle 101, and improves reliability and safety of a self-driving technology or the like.

[0026] Before the method for determining a vehicle speed provided in the embodiments of this application is described, a hardware architecture of a vehicle in the embodiments of this application is first described.

[0027] FIG. 2 is a schematic diagram of a hardware architecture of a system applied to a vehicle according to an embodiment of this application. The vehicle 200 includes a front-view camera 201, a radar 202, a global positioning system (English: Global Positioning System, GPS for short) 203, an image processing unit 204, a central processing unit (English: Central Processing Unit, CPU for short) 205, and a controller 206. The front-view camera 201 may be configured to collect an image of a road scenario. The radar 202 may be configured to collect data of a dynamic surrounding object or a static surrounding object. The image processing unit 204 may be configured to recognize a lane line, a lane curb, another vehicle, and a surrounding object (for example, a pedestrian, an animal, or a tree). The CPU 205 may be configured to perform overall control on the entire vehicle 200 by performing the following operations: obtaining image data from the front-view camera 201 and status data of the surrounding object from the radar 202; invoking the image processor 204 to perform target recognition and invoking an internal computation module of the CPU 205 to perform fusion and other operations; determining an appropriate target vehicle speed, and generating a decision control instruction based on the target vehicle speed; and sending the decision control instruction to the controller 206. The controller 206 may be configured to control, based on the received decision control instruction, the vehicle to travel in a current lane at the target speed.

[0028] For the vehicle 200 of the hardware architecture shown in FIG. 2, a schematic diagram of a corresponding system architecture is shown in FIG. 3 according to an embodiment of this application. From a perspective of a system, the vehicle 200 includes an in-vehicle sensor system 210, an in-vehicle computer system 220, and an in-vehicle control execution system 230. The in-vehicle sensor system 210 may be configured to obtain data collected by the front-view camera 201, data collected by the radar 202, and data obtained through positioning by the GPS 203. The in-vehicle computer system 220 is generally divided into two modules: a perception data processing module 221 and a decision-making and planning module 222. The perception data processing module 221 may be configured to: detect a surrounding object (especially a surrounding pedestrian) of the vehicle 200, and output a position and motion information of the surrounding object. The decision-making and planning module 222 may be configured to predict and update a distribution of intentions of the surrounding object based on a current position and the motion information of the surrounding object, so as to decide and plan a vehicle speed of the vehicle 200 based on the distribution of intentions. The in-vehicle control execution system 230 may be configured to: obtain a decision control instruction output by the decision-making and planning module 222, and control the vehicle 200 to travel at a vehicle speed indicated in the decision control instruction. It should be noted that the method for determining a vehicle speed provided in this embodiment of this application is mainly performed by the decision-making and planning module 222 of in in-vehicle computer system 220. For a specific implementation, refer to related descriptions in the embodiment shown in FIG. 4.

[0029] In an example, from a perspective of a product, a corresponding schematic structural diagram of the vehicle 200 in this embodiment of this application is shown in FIG. 4. The vehicle 200 includes a sensor layer 410, a perception layer 420, a decision-making and planning layer 430, and a vehicle control layer 440. A data stream sequentially arrives at the foregoing four layers, and is sequentially processed by the four layers. The sensor layer 410 may be configured to load the data collected by the monocular/binocular front-view camera 201, the data collected by the lidar/millimeter-wave radar 202, and the data obtained through positioning by the GPS 203. The perception layer 420 may be configured to load data obtained by the following six modules: a vehicle/surrounding object detection module 421, a lane line detection module 422, a traffic sign detection module 423, an ego vehicle positioning module 424, a dynamic/static object detection module 425, and a perception and fusion module 426. The decision-making and planning layer 430 may be configured to load data obtained by a pedestrian intention distribution prediction and update module 431, a speed decision-making and planning module 432, and a path planning module 433. The vehicle control layer 440 may be configured to perform horizontal and vertical control on the vehicle 200 based on data sent by the decision-making and planning layer 430. It should be noted that a module in a gray box in FIG. 4 is a module configured to implement the method for determining a vehicle speed provided in this embodiment of this application. In this embodiment of this application, whether a determined speed of the vehicle is safe and reliable mainly depends on the two modules 431 and 432 in gray in the decision-making and planning layer 430.

[0030] It may be understood that the foregoing scenario is merely an example of a scenario provided in this embodiment of this application, and this embodiment of this application is not limited to this scenario.

[0031] With reference to the accompanying drawings, the following describes in detail a specific implementation of a method for determining a vehicle speed in the embodiments of this application by using embodiments.

[0032] FIG. 5 is a schematic flowchart of a method for determining a vehicle speed according to an embodiment of this application. The method may specifically include the following step 501 to step 505.

[0033] Step 501: Obtain observation information of a surrounding object of a vehicle.

[0034] Step 502: Compute, based on the observation information of the surrounding object, a probability distribution of different intentions of the surrounding object.

[0035] It may be understood that, in a driving environment of the vehicle, the surrounding object of the vehicle may include an object, for example, a pedestrian or an animal around the vehicle, that may participate in traffic. In this embodiment of this application, the surrounding object of the vehicle may be understood and described by using a pedestrian around the vehicle as an example. The observation information of the surrounding object is information that can reflect a status of the surrounding object, and can be used to predict a probability of each intention of the surrounding object. It may be understood that the intention is an intention of the surrounding object relative to a current lane. For example, FIG. 6 is a schematic diagram of a plurality of intentions of a pedestrian. It can be learned that, in FIG. 6, diagram a shows an intention g1: stepping forward along the sidewalk; diagram b shows an intention g2: stepping backward along the sidewalk; diagram c shows an intention g3: straight crossing the lane; diagram d shows an intention g4: diagonally crossing the lane in a driving direction; diagram e shows an intention g5: stepping away from the lane; diagram f shows an intention g6: diagonally crossing the lane in a direction opposite to a driving direction; and diagram g shows an intention g7: stop. In one case, if there is only one surrounding object around the vehicle, an intention of a surrounding object means an intention of the only one surrounding object. For example, it is assumed that the surrounding object of the vehicle is a pedestrian A. Assuming that each pedestrian has two possible intentions: waiting and crossing, the intention of the surrounding object includes two intentions: Awaits and A crosses the lane. In another case, if there are at least two surrounding objects around the vehicle, an intention of a surrounding object means a combination of intentions corresponding to each of the surrounding objects. For example, it is assumed that surrounding objects of the vehicle include a pedestrian A and a pedestrian B. Assuming that each pedestrian has two possible intentions: waiting and crossing, intentions of the surrounding objects include 2 x 2 = 4 intention combinations: {A waits, B waits}, {A waits, B crosses the lane}, {A crosses the lane, B waits}, and {A crosses the lane, B crosses the lane}.

[0036] During specific implementation, a probability of each intention of the surrounding object may be computed based on the obtained observation information of the surrounding object of the vehicle. The occurrence probability of each intention of the surrounding object can be used to obtain the probability distribution of different intentions of the surrounding object. The occurrence probability is a probability that each surrounding object has each intention.

[0037] In an example, step 502 may specifically include: establishing, based on the observation information of the surrounding object by establishing a coordinate system related to the lane in which the vehicle travels, a relative position relationship between the surrounding object and the lane and a relative motion relationship between the surrounding object and the lane; and computing the probability distribution of the different intentions of the surrounding object based on the relative position relationship between the surrounding object and the lane and the relative motion relationship between the surrounding object and the lane.

[0038] It may be understood that, in the lane coordinate system (namely, an S-L coordinate system), a direction that uses a start point of the path as an origin and in which the vehicle is to travel in the lane is marked as a positive direction of an S axis, and a left direction perpendicular to the positive direction of the S axis is a positive direction of an L axis. For details, reference may be made to FIG. 7.

[0039] In some implementations, in step 501 and step 502, the occurrence probability of each intention of the surrounding object may be specifically predicted based on observation information of the surrounding object at a previous moment and observation information of the surrounding object at a next moment. Specific implementation may include: S11: Obtain the observation information of the surrounding object of the vehicle. S12: Determine whether each surrounding object is a new surrounding object; if the surrounding object is a new surrounding object, perform S13; if the surrounding object is not a new surrounding object, perform S14. S13: Initialize an occurrence probability of each intention of the surrounding object. S14: Update an occurrence probability of each intention of the surrounding object based on the observation information. It should be noted that, after the occurrence probability of each intention of the surrounding object is determined, a probability distribution of different intentions of the surrounding object may be determined based on the occurrence probability of each intention.

[0040] It should be noted that the observation information, obtained in step 501 (namely, S11), of the surrounding object of the vehicle may be specifically obtained in the following process: The front-view camera 201, the radar 202, or the GPS 203 in FIG. 2, FIG. 3, or FIG. 4 collects the observation information. The observation information is sent to the vehicle/surrounding object detection module 421, the ego vehicle positioning module 424, and the dynamic/static object detection module 425 of the perception layer 420 in FIG. 4 for separate processing. Then, three processing results are sent to the perception and fusion module 426 for data association fusion and tracking processing. Step 502 (namely, S12 to S14) may be specifically implemented by the CPU 205 (the decision-making and planning module 222 in the in-vehicle computer system 220 in FIG. 3 or the pedestrian intention distribution prediction and update module 431 in the decision-making and planning layer 430 in FIG. 4) in FIG. 2.

[0041] In an example, S11 may specifically include: obtaining first observation information of the surrounding object in the rectangular coordinate system by performing processing such as filtering, multi-sensor data association and fusion, and tracking on collected data of an environment around the vehicle. The first observation information of the surrounding object may include a position of the surrounding object, a motion speed of the surrounding object, and a motion heading of the surrounding object. It should be noted that, for each surrounding object participating in traffic, first observation information of the surrounding object needs to be obtained. To provide a data foundation for subsequent computation, first observation information of the ego vehicle also needs to be obtained, including a vehicle position, a vehicle speed, a vehicle acceleration, and a vehicle course.

[0042] The traffic scenario shown in FIG. 1 is still used as an example. First observation information of the pedestrian 102 may be obtained through S11 and may be represented as: Opedestrian 102={position: (xpedestrian 102, ypedestrian 102), motion speed: Vpedestrian 102, motion heading: θpedestrian 102}. First observation information of the pedestrian 103 may be represented as: Opedestrian 103={position: (Xpedestrian 103, ypedestrian 103), motion speed: Vpedestrian 103, motion heading: θpedestrian 103}. First observation information of the vehicle 101 may be represented as: Ovehicle 101={position: (Xvehicle 101, yvehicle 101), speed: Vvehicle 101, acceleration: avehicle 101, course: θvehicle 101}.

[0043] During specific implementation, to consider, from a perspective of the vehicle, whether the surrounding object has a possibility of entering the lane in which the vehicle is to travel, the surrounding object needs to be observed in the S-L coordinate system. In this case, a position in the first observation information needs to be transformed from the rectangular coordinate system to a position, in the S-L coordinate system, that is used as a position in second observation information. A specific transformation may include: vertically mapping the original position in the rectangular coordinate system to a mapping point in a direction of the lane in which the vehicle is to travel; obtaining a distance between the start point of the driving lane and the mapping point, and using the distance as a value in the S axis direction; and computing a distance between the original position and the mapping point, and using the distance as a value in the L axis direction. Referring to FIG. 7, a vehicle-surrounding object-lane model may be constructed. The vehicle-surrounding object-lane model uses the S-L coordinate system as a reference coordinate system, and is used to describe a relative position relationship between the vehicle and the lane and a relative motion relationship between the vehicle and the lane, and a relative position relationship between the surrounding object and the lane and a relative motion relationship between the surrounding object and the lane. Therefore, the vehicle-surrounding object-road model may be used to compute the second observation information. For example, for a pedestrian and a vehicle shown in FIG. 7, it is assumed that first observation information of the pedestrian includes a position (xpedestrian, ypedestrian), and first observation information of the vehicle includes a position (xvehicle, yvehicle). In this case, referring to FIG. 7, a position, in the S-L coordinate system, to which the pedestrian position is transformed is (spedestrian, lpedestrian), and a position, in the S-L coordinate system, to which the vehicle position is transformed is (svehicle, lvehicle).

[0044] The traffic scenario shown in FIG. 1 is still used as an example. Second observation information of the pedestrian 102 may be obtained through S 11 and may be represented as: Opedestrian 102'={(Spedestrian 102, lpedestrian 102), Vpedestrian 102, θpedestrian 102}. Second observation information of the pedestrian 103 may be represented as: Opedestrian 103'={(Spedestrian 103, lpedestrian 103), Vpedestrian 103, θpedestrian 103}. Second observation information of the vehicle 101 may be represented as: Ovehicle 101'={(Svehicle 101, lvehicle 101), Vvehicle 101, avehicle 101, θvehicle 101}.

[0045] It should be noted that, after the observation information of the surrounding object of the vehicle is obtained according to the foregoing implementation, S12 may be performed to determine whether each surrounding object is a new surrounding object. For the new surrounding object, an occurrence probability of each intention of the surrounding object is initialized according to S13. For an existing surrounding object, an occurrence probability of each intention of the surrounding object is updated based on the observation information according to S14. It may be understood that in both S13 and S14, the probability is obtained through computation based on the relative position relationship between the surrounding object and the lane and the relative motion relationship between the surrounding object and the lane.

[0046] For S12, whether the surrounding object is a new surrounding object may be determined by determining whether the surrounding object observed at a current moment has been observed before the current moment. If the surrounding object observed at the current moment has not been observed before the current moment, it indicates that the surrounding object is a new object that appears around the vehicle, and it may be determined that the surrounding object is a new surrounding object. On the contrary, if the surrounding object observed at the current moment has also been observed before the current moment, it indicates that the surrounding object exists before the current moment, and it may be determined that the surrounding object is not a new surrounding object.

[0047] In S13, an occurrence probability of a new surrounding object is initialized. Because the surrounding object is newly observed, and there is no other data foundation of an intention of the surrounding object, the occurrence probability may be determined based on a quantity of possible intentions of the surrounding object. In other words, the occurrence probability of each possible intention is equal. For example, assuming that a new surrounding object A has seven possible intentions, occurrence probabilities of all of the seven predicted intentions are equal, that is, 1/7.

[0048] For S14, the occurrence probability of each intention of the surrounding object is updated based on the observation information. Specifically, the occurrence probability of each intention of the surrounding object may be determined based on an occurrence probability of the intention at a moment closest to the current moment, a position of the surrounding object at the current moment and a position of the surrounding object at the moment closest to the current moment, and means and corresponding variances of an update model of the intention in the S direction and the L direction in the S-L coordinate system. For example, the update model may be a Gaussian motion model.

[0049] For example, it is assumed that only the pedestrian 102 can be observed at a moment when t=0, and the pedestrian 102 and the pedestrian 103 can be observed at a moment when t=1.

[0050] At the moment when t=0, the observation information of the pedestrian 102 in the rectangular coordinate system is obtained as

. If the Pedestrian 102 is determined, through S12, as a new surrounding object, a status of the pedestrian 102 and a distribution of occurrence probabilities of intentions of the pedestrian 102 need to be initialized. A specific operation may include: using a path path = {(x0, y0),(x1, y1), (x2, y2),..., (xn, yn)} on which the vehicle is to travel as a reference coordinate system (where n is a quantity of points included in the reference coordinate system); computing a point (xi, yi) of a position

of the pedestrian 102 projected on the path; and computing a distance

, along the path, between a start point (x0, y0) of the path and the projected point (xi, yi), and computing a distance

between

and the projected point (xi,yi). In this case, an initialized status of the pedestrian 102 may be

, where

is the distribution of occurrence probabilities of the plurality of intentions of the pedestrian 102 at the moment when t=0. Because the pedestrian 102 is a new surrounding object, it is determined that

. The intentions of the pedestrian 102 include seven intentions: g1 to g7, and probabilities of all the intentions are equal (that is, 1/7). It may be understood that

represents a probability that the pedestrian 102 has the intention g1 at the moment t=0,

represents a probability that the pedestrian 102 has the intention g2 at the moment t=0, and so on. Details are not described herein.

[0051] At the moment t=1, the observation information of the pedestrian 102 and the observation information of the pedestrian 103 in the rectangular coordinate system are respectively

and

After determining in S12, if it is determined that the pedestrian 103 is a new surrounding object, a status of the pedestrian 103 and a distribution of intentions of the pedestrian 103 need to be initialized; and if it is determined that the pedestrian 102 is an existing surrounding object, an occurrence probability of each intention of the pedestrian 102 is updated based on the observation information. For a process of initializing the status and the distribution of intentions of the pedestrian 103, refer to the description of the initialization process for the pedestrian 102 at the moment t=0. Details are not described herein again. A process of updating each intention of the pedestrian 102 may specifically include the following steps: Step 1: Obtain a position and a speed





of the pedestrian 102 in the S-L coordinate system at the moment t= 1 through vertical projection and distance computation based on the observation information

of the pedestrian 102. Step 2: Update a distribution

of intentions of the pedestrian 102 at the moment t=1 according to the Gaussian motion model. Specifically, an occurrence probability of each intention is updated. For step 2, the occurrence probability of the intention g1 at the moment t=1 is used as an example for description. Specifically, an occurrence probability of the intention g1 at the moment t=1 after the update may be computed according to the following formula (1):

where

and

may be obtained through computation according to the following formula (2):

where

and

are means of the intention g1 of the pedestrian 102 in the S direction and the L direction in the S-L coordinate system by using the Gaussian motion model. σ8 and σl are standard deviations of the intention g1 of the pedestrian 102 in the S direction and the L direction in the S-L coordinate system by using the Gaussian motion model.

[0052] It may be understood that a manner of updating occurrence probabilities of the intentions g2 to g7 of the pedestrian 102 at the moment t=1 is similar to the manner of updating the occurrence probability of the intention g1 at the moment t=1. Details are not described herein again. After the update, at the moment t=1, a distribution of occurrence probabilities of the intentions of the pedestrian 102 may be represented as

It may be understood that,

represents a probability that the pedestrian 102 has the intention g1 at the moment t=1,

represents a probability that the pedestrian 102 has the intention g2 at the moment t=1, and so on. Details are not described herein.

[0053] It should be noted that, in a manner of computing the distribution of occurrence probabilities of the intentions of the pedestrian 102, to cover more surrounding objects and comprehensively compute an occurrence probability of each possible intention of each surrounding object, a concept of a particle may be introduced. An occurrence probability of each intention is represented by a quantity of particles included in the intention. Specifically, if the intention g1 includes a large quantity of particles, it indicates that the occurrence probability of the intention is relatively high. On the contrary, if the intention g1 includes a small quantity of particles, it indicates that the occurrence probability of the intention is relatively low.

[0054] For example, in the corresponding example of S13 in the foregoing first implementation, the distribution of occurrence probabilities of the intentions is initialized as

at the moment t=0. In this case, the distribution of occurrence probabilities may be represented by using particles of a preset quantity (namely, an integer multiple of a quantity of all intentions, for example, 700). As shown in FIG. 8, each intention corresponds to a particle set that includes a same quantity of particles. For example, for 700 particles that include seven intentions, an occurrence probability of each intention is 1/7. Therefore, each intention corresponds to a set of 100 same particles, and a weight of each particle is 1/700.

[0055] A status of one of the 700 particles may be represented as particle = {svehicle, lvehicle, vvehicle, spedestrian j, lpedestrian j, vpedestrian j, gpedestrian j, w} , where j represents a jth surrounding object of the vehicle, and w represents a weight of the particle. A set of particles may be expressed as {particle1, particle2, ..., particlem}, where m is a total quantity of particles, namely, 700. For example, for a vehicle and surrounding objects: a pedestrian 1 and a pedestrian 2 of the vehicle, a status of each particle may be represented as

, where wi represents a weight of a particlei, and the weight is used to represent a risk degree of an intention corresponding to the particle. For specific descriptions, refer to related descriptions in the following step 503 to step 505.

[0056] In some other implementations, in step 501 and step 502, specifically, a probability distribution of intentions of the surrounding object may be output based on the observation information of the current surrounding object by using a trained machine learning model. The observation information of the surrounding object may be observation information that is specifically processed in the foregoing implementation and that includes a position, a motion speed, and a motion heading of the surrounding object, or may be a currently collected image including the surrounding object.

[0057] In this implementation, in a case, if the observation information is the foregoing processed observation information that includes the position, the motion speed, and the motion heading of the surrounding object, a first machine learning model may be trained and pre-constructed based on a large amount of historical observation information corresponding to a learned occurrence probability of each intention, and the corresponding learned occurrence probability of each intention, to obtain a first machine learning model that has been trained. Then, the observation information that is of the surrounding object and that is obtained in step 501 may be input into the first machine learning model that has been trained, and an occurrence probability of each intention of the surrounding object is output.

[0058] In another case, if the observation information is a currently collected image including the surrounding object, a second machine learning model may be trained and pre-constructed based on a large quantity of historical images corresponding to a learned occurrence probability of each intention, and the corresponding learned occurrence probability of each intention, to obtain a second machine learning model that has been trained. Then, the observation information (namely, the currently collected image including the surrounding object) that is of the surrounding object and that is obtained in step 501 may be input into the second machine learning model that has been trained, and an occurrence probability of each intention of the surrounding object included in the image is output.

[0059] It may be understood that, in both of the foregoing two implementations, an occurrence probability of each of a plurality of intentions of the surrounding object can be predicted based on the observation information of the surrounding object. This provides an indispensable data foundation for subsequent accurate determining of a vehicle speed of the vehicle, and therefore improves safety and reliability of the vehicle that uses intelligent driving technologies such as a self-driving technology.

[0060] Step 503: Perform redistribution computation on the probability distribution based on travel times required for the vehicle to travel from a current position of the vehicle to risk areas corresponding to the different intentions, to obtain a probability redistribution of the different intentions, where the risk areas corresponding to the different intentions are areas, in the lane in which the vehicle travels, located proximate to the surrounding object with the different intentions.

[0061] Step 504: Predict motion status variations of the surrounding object with the different intentions based on the travel times required for the vehicle to travel to the risk areas corresponding to the different intentions.

[0062] It may be understood that, to ensure that a determined vehicle speed is more secure and reliable, the vehicle speed of the vehicle needs to be determined based on at least the probability redistribution of the intentions of the surrounding object and the motion status variations of the surrounding object with the different intentions. Both the probability redistribution of the intentions of the surrounding object and the motion status variations of the surrounding object need to be obtained through computation based on the travel times required for the vehicle to travel to the risk areas corresponding to different intentions of the surrounding object. The travel time is used to quantize a risk degree of each intention, namely, a possibility of a collision with the vehicle when the rounding object moves with the intention. It may be understood that a time to collision under each intention is the travel time required for the vehicle to travel to the risk area corresponding to each intention of the surrounding object. For example, as shown in FIG. 9, for the intention g3 (namely, straight crossing the lane) of a pedestrian, a risk area is an area A, in the lane in which the vehicle travels, located proximate to the pedestrian with the intention of straight crossing the lane. In this case, a corresponding time to collision is a travel time ttcg3 required for the vehicle to travel from a current position to the area A. For the intention g4 (namely, diagonally crossing the lane in a driving direction) of the pedestrian, a risk area is an area B, in the lane in which the vehicle travels, located proximate to the pedestrian with the intention of straight crossing the lane. In this case, a corresponding time to collision is a travel time ttcg4 required for the vehicle to travel from the current position to the area B.

[0063] In some implementations, before step 503, in this embodiment of this application, the travel times required for the vehicle to travel from the current position of the vehicle to the risk areas corresponding to the different intentions may be further computed in step 501 and the following S21 to S24: S21: Obtain the observation information of the vehicle. S22: Establish, based on the observation information of the vehicle and the observation information of the surrounding object by establishing a coordinate system related to the lane in which the vehicle travels, a relative position relationship between the vehicle and the lane and a relative motion status between the vehicle and the lane, and a relative position relationship between the surrounding object and the lane and a relative motion status between the surrounding object and the lane. S23: Determine, based on the relative position relationship between the surrounding object and the lane and the relative motion status between the surrounding object and the lane, the risk areas corresponding to the different intentions. S24: Compute, based on the relative position relationship between the vehicle and the lane and the relative motion status between the vehicle and the lane, and the risk areas corresponding to the different intentions, the travel times required for the vehicle to travel from the current position of the vehicle to the risk areas corresponding to the different intentions. For specific implementation of S21, refer to the related descriptions of obtaining the observation information of the surrounding object in step 501. For specific implementation of S22, refer to the related descriptions of transformation between coordinate systems in S11.

[0064] It may be understood that, for an intention, if a time to collision under the intention is long, it indicates that a probability of a risk that occurs under the intention is low, that is, a risk degree is low. On the contrary, if a time to collision under the intention is short, it indicates that a probability of a risk occurred under the intention is relatively high, that is, a risk degree is high. For example, for the pedestrian in FIG. 9, ttcg4 is clearly greater than ttcg3, indicating that if the pedestrian straight crosses the lane, there is a high probability that the pedestrian collides with the vehicle, and a risk degree is high. However, if the pedestrian diagonally crosses the lane in a driving direction, compared with the high collision probability under the intention of straight crossing the lane, a possibility of collision with the vehicle is reduced, and a risk degree is reduced.

[0065] In an example, after the time to collision is determined, step 503 may be specifically implemented through the following S31 to S32: S31: Perform particle processing on the probability distribution, where quantities of particles corresponding to the different intentions are used to represent the probability distribution of the different intentions. S32: Adjust, based on the travel times that are obtained through computation and that are required for the vehicle to travel to the risk areas corresponding to the different intentions, weights of the particles corresponding to the different intentions, to obtain the probability redistribution of the different intentions.

[0066] For details of S31, refer to FIG. 8 and the related descriptions in the foregoing embodiment. Details are not described herein again.

[0067] For computation of the travel times required for the vehicle to travel to the risk areas corresponding to the different intentions in S32, refer to the foregoing descriptions of related parts in S21 to S24. A specific principle is not described herein again. For clearer description, the following uses an example to describe an example process of computing the travel time after particle processing. For example, it is assumed that for the vehicle and the pedestrian 1 and the pedestrian 2 around the vehicle, for a particle i,

Under a same intention

a risk area of the pedestrian 1 is determined, and a travel time

required for the vehicle to travel to the risk area corresponding to the intention of the pedestrian 1 is computed; and a risk area of the pedestrian 2 is determined, and a travel time

required for the vehicle to travel to the risk area corresponding to the intention of the pedestrian 2 is computed. To maximally reduce a collision possibility for the pedestrian 1 and the pedestrian 2, a smaller travel time in travel times corresponding to pedestrians is selected as a travel time ttci of the particle i, that is,

.

[0068] In S32, the weights of the particles corresponding to the different intentions are adjusted based on the travel times that are obtained through computation and that are required for the vehicle to travel to the risk areas corresponding to the different intentions, to obtain the probability redistribution of the different intentions. A travel time (namely, a time to collision) of each particle represents a risk degree of a collision of the surrounding object with a specific intention, and a shorter travel time indicates a higher risk degree. Therefore, to increase attention to an intention of a high risk degree, a weight of a particle with a high risk degree may be increased based on the travel time according to the following formula (3):

where W represents a risk coefficient, and ε represents an effective computation constant. A smaller travel time ttci indicates a larger weight of a particle, obtained through computation, that indicates a risk degree of the particle. In this way, the risk degree of the particle can be highlighted. In addition, in order that computation can be converged, normalization processing may be further performed on the weight

. Specifically, a weight

of the particle i may be computed according to the following formula (4):



[0069] It may be understood that, according to the foregoing description, a risk degree of each intention can be determined based on the travel time required for the vehicle to travel to the risk area corresponding to each intention. In other words, the probability redistribution of the different intentions can be implemented. This provides an indispensable data foundation for subsequent accurate determining of the vehicle speed of the vehicle, and therefore improves safety and reliability of the vehicle that uses intelligent driving technologies such as a self-driving technology.

[0070] In an example, in step 504, specifically, the motion status variations of the surrounding object with the different intentions may be predicted through the following S41 and S42: S41: Determine, based on the travel times required for the vehicle to travel to the risk areas corresponding to the different intentions, probabilities that the surrounding object with the different intentions changes the current intention. S42: Predict the motion status variations of the surrounding object with the different intentions based on the probabilities that the surrounding object with the different intentions changes the current intention and a random probability.

[0071] It may be understood that, in S41, an interaction probability related to the vehicle and the surrounding object may be further determined based on a time to collision ttc under each intention. In one case, if an interaction probability under an intention of the surrounding object is excessively high, the current intention may be changed, based on the interaction probability, to a target intention. The target intention is an intention of the surrounding object after adjustment. For example, if an interaction probability corresponding to a time to collision ttc under the intention g1 of the pedestrian 1 is quite high, the intention g2 may be determined as the target intention of the pedestrian 1 based on the interaction probability. In another case, if an interaction probability under an intention of the surrounding object is relatively low, it may be determined, based on the interaction probability, that the current intention is still the target intention. The target intention is the intention of the surrounding object before adjustment. For example, if an interaction probability corresponding to a time to collision ttc under the intention g1 of the pedestrian 1 is quite low, the intention g1 may be determined as the target intention of the pedestrian 1 based on the interaction probability.

[0072] It may be understood that, if a time to collision between the surrounding object and the vehicle is relatively short when the surrounding object has a specific intention, that is, a risk degree is relatively high, the surrounding object is generally cautious. On the contrary, if a time to collision between the surrounding object and the vehicle is relatively long when the surrounding object has a specific intention, that is, a risk degree is relatively low, the surrounding object is generally relaxed. In view of this, the interaction probability obtained through computation based on the time to collision ttc is introduced, so that a motion status that matches psychology of the pedestrian can be realistically simulated. During specific implementation, it is assumed that a time to collision between the pedestrian 1 and the vehicle is ttc pedestrian 1, and a time to collision between the pedestrian 2 and the vehicle is

. For example, an interaction probability related to the pedestrian 1 and the vehicle may be: Pr(pedestrian 1^vehicle) =

. Likewise, an interaction probability related to the pedestrian 2 and the vehicle may be

, where Winteract is an interaction probability coefficient.

[0073] During specific implementation, in S42, whether to use a surrounding object interactive motion model or a surrounding object linear motion model is determined based on a surrounding object status prediction model and the computed interaction probability, to determine the motion status variation of the surrounding object. The pedestrian behaves at random and an interaction probability related to the pedestrian and the vehicle is also random. Therefore, a random probability Prandom is introduced to determine a model that is supposed to be used to compute an initial expectation value. A specific determining process includes the following steps: Step 1: Determine whether the interaction probability Pr is greater than the random probability Prandom. If the interaction probability Pr is greater than the random probability Prandom, perform step 2; or if the interaction probability Pr is not greater than the random probability Prandom, perform step 3. Step 2: Predict the motion status variations of the surrounding object with the different intentions by using the surrounding object interactive motion model. Step 3: Predict the motion status variations of the surrounding object with the different intentions by using the surrounding object linear motion model.

[0074] A scenario in FIG. 10 is used as an example for description. It is assumed that an intention of a pedestrian 1 is g3: straight crossing the lane. If the surrounding object interactive motion model is not used, the pedestrian 1 moves to a position ② in FIG. 12, ignoring existence of the vehicle. If the surrounding object interactive motion model is used, the pedestrian 1 is very likely to move to a position ① to avoid the vehicle for safety. Similarly, because a time to collision between a pedestrian 2 and the vehicle is relatively long, and an interaction probability related to the pedestrian 2 and the vehicle is relatively low, a possibility that the pedestrian 2 moves to a position ④ in use of the surrounding object linear motion model is greater than a possibility that the pedestrian 2 moves to a position ③ in use of the surrounding object interactive motion model.

[0075] For the surrounding object linear motion model, an error of the observation information of the position and the speed of the surrounding object is relatively large. Therefore, in the surrounding object linear motion model, motion statuses of the surrounding object may be set to a Gaussian distribution with a relatively large variance. The surrounding object linear motion model is specifically defined as follows:

Δt represents a predicted time step, and is usually relatively small. For example, Δt may be set to 0.3 second. It is assumed that the intention of the pedestrian remains unchanged within Δt, that is,

. In addition, ƒs(gpedestrian) and ƒl(gpedestrian) respectively represent motion heading components of different intentions in the S direction and the L direction in the S-L coordinate system. In other words, motion headings of the surrounding object with different intentions are different. µpedestrian s and µpedestrian l respectively represent means of motion distances of the surrounding object linear motion model in the S direction and the L direction;

and

respectively represent variances of the motion distances of the surrounding object linear motion model in the S direction and the L direction; and µpedestrian v and

respectively represent a mean of motion speeds of the surrounding object linear motion model in the S direction and a variance of the motion speeds of the surrounding object linear motion model in the L direction.

[0076] A specific definition of the surrounding object interactive motion model is as follows:

Fs(νpedestriant,gpedestrian) and Fl(νpedestriant,gpedestrian) respectively represent motion variation functions of the pedestrian in the S direction and the L direction in the S-L coordinate system when the pedestrian with different intentions interacts with the vehicle.

[0077] It should be noted that step 503 and step 504 are not sequential. Step 503 may be performed before or after step 504, or step 503 and step 504 may be simultaneously performed. A specific manner is not limited.

[0078] Step 505: Determine a travelling speed of the vehicle based on the probability redistribution of the different intentions, the motion status variations of the surrounding object with the different intentions, and motion status variations of the vehicle under different travelling speed control actions.

[0079] It may be understood that an appropriate vehicle speed of the vehicle may be determined based on three factors: the probability redistribution of the different intentions, the motion status variations of the surrounding object with the different intentions, and the motion status variations of the vehicle under different travelling speed control actions. In one case, an acceleration of the vehicle may be determined based on the foregoing three factors, and the vehicle is controlled to travel at the acceleration. In another case, an acceleration of the vehicle may be determined based on the foregoing three factors, so that a travelling speed at which the vehicle is to travel is determined based on the acceleration and a current speed of the vehicle, and the vehicle is controlled to travel at the determined travelling speed.

[0080] During specific implementation, step 505 may be specifically implemented through the following S51 to S53: S51: Estimate travelling effects of the vehicle that bring about under the different travelling speed control actions based on the probability redistribution of the different intentions, the motion status variations of the surrounding object with the different intentions, and the motion status variations of the vehicle under the different travelling speed control actions. S52: Select a target travelling speed control action from the different travelling speed control actions based on the travelling effects of the vehicle that bring about under the different travelling speed control actions. S53: Determine the travelling speed of the vehicle based on the target travelling speed control action.

[0081] Specifically, in S51, a vehicle status prediction model may be established, and the travelling effects, namely, motion status variations of the vehicle when the vehicle travels at different accelerations, of the vehicle under different travelling speed control, may be predicted based on the vehicle status prediction model.

[0082] For the vehicle status prediction model, an error in observation information of status parameters such as a position and a speed of the vehicle is relatively small. Therefore, in the vehicle status prediction model, motion statuses of the vehicle may be set to a Gaussian distribution with a relatively small variance. The vehicle status prediction model is specifically defined as follows:

µvehicle s and µvehicle l respectively represent means of motion distances of the vehicle status prediction model in the S direction and the L direction;

and

respectively represent variances of the motion distances of the vehicle status prediction model in the S direction and the L direction; and µvehicle v and

respectively represent a mean of motion speeds of the vehicle status prediction model in the S direction and a variance of motion speeds of the vehicle status prediction model in the L direction.

[0083] It may be understood that, considering that an intention of the surrounding object is not definite, a partially observable Markov decision process (English: Partially Observable Markov Decision Process, POMDP for short) may be used to perform decision-making and planning with an optimal speed. It may be understood that the POMDP has a feature of partial observing. To be specific, an intention of an unobservable part in an uncertain environment is predicted after decision-making and planning is performed by using a general mathematical model. The mathematical model may generally include a state set S, an action set A, a state transition function T, an observation set O, an observation function Z, and a reward function R. With reference to the scenario corresponding to FIG. 1, content included in the mathematical model is defined as follows:
State space S is a set of all possible states of a dynamic entity and a static entity in an environment, namely, the vehicle, a pedestrian 1 (namely, the pedestrian 102 in the foregoing description), and a pedestrian 2 (that is, the pedestrian 103 in the foregoing description). S = {s|s ∈ [svehicle, lvehicle, vvehicle, spedestrian 1, lpedestrian 1, Vpedestrian 1, gpedestrian 1, spedestrian 2, lpedestrian 2, Vpedestrian 2, gpedestrian 2]}.

[0084] Action space A is a set of acceleration actions that may be used by a self-driving or unmanned vehicle. For ease of description, an extracted common acceleration range is usually discretized, or may also be understood as corresponding gears, for example, A={-3, -2, -1, 0, 0.5, 1, 2, 3}. In other words, the vehicle may travel at eight different initial accelerations.

[0085] The state transition function (T) is a critical part of POMDP. This function T describes a state transition process over time and provides a decision basis for selection of an optimal action. For the vehicle, the state transition function T may indicate that the vehicle transits to a

state after using an acceleration a in A in a state {svehicle, lvehicle, νvehicle}. For the pedestrian 1, the state transition function T indicates that the pedestrian 1 transits to a state

when moving under an intention gpedestrian 1 in a Current state {spedestrian 1, lpedestrian 1, Vpedestrian 1, gpedestrian 1}.

[0086] The observation space O generally corresponds to the state space, and represents an observation information set of the vehicle, the pedestrian 1, and the pedestrian 2, that is, 0 = {o|o e [ovehicle, opedestrian 1, opedestrian 2]}. Ovehicle={position: (xvehicle, yvehicle), speed: Vvehicle, acceleration: avehicle, course: θvehicle}. Opedestrian 1={position: (Xpedestrian 1, ypedestrian 1), motion speed: Vpedestrian 1, motion heading: θpedestrian 1}. Opedestrian 2={position: (Xpedestrian 2, ypedestrian 2), motion speed: Vpedestrian 2, motion heading: θpedestrian 2}.

[0087] The observation function Z represents a probability of obtaining the observation function z after the vehicle, the pedestrian 1, and the pedestrian 2 transit to a state s' after using the acceleration a, that is, Z(z,s',a) = P(z|s',a). It is assumed that positions and speeds of the vehicle and the pedestrian match a Gaussian distribution relative to actual positions and speeds. Because the error of the observation information of the position and the speed of the vehicle is relatively small, and the error of the observation information of the position and the speed of the pedestrian is relatively large, a variance of a Gaussian distribution for the vehicle is different from a variance of a Gaussian distribution for the pedestrian. A variance of a Gaussian distribution of a Gaussian motion model of the vehicle is relatively small, and a variance of a Gaussian distribution of a motion model of the pedestrian is relatively large.

[0088] The reward function Reward is used to perform quantitative assessment on the determined acceleration. The assessment may be performed based on a collision degree, or a collision degree and a traffic obstruction degree, or a collision degree and a travelling discomfort degree, or a collision degree, a traffic obstruction degree, and a travelling discomfort degree. The collision degree reflects safety, the traffic obstruction degree reflects traffic efficiency, and the travelling discomfort degree may reflect comfort. It should be noted that the determined acceleration may alternatively be assessed based on a purpose.

[0089] For example, if the determined acceleration is assessed based only on the collision degree R_col, Reward=R_col. For another example, if the determined acceleration is assessed based on the collision degree R col, the traffic obstruction degree R move, and the travelling discomfort degree R_action, Reward = R col + R_move + R_action.

[0090] After some definitions in POMDP are described, the following describes an example specific implementation of S51 in step 505.

[0091] It may be understood that all possible accelerations of the vehicle may be traversed, and [svehicle', lvehicle', vvehicle'] that dynamically varies is predicted correspondingly by using the vehicle status prediction model, and is compared with [spedestrian', lpedestrian', vpedestrian', gpedestrian'], to determine whether the vehicle collides with the surrounding object. If no collision occurs, it is determined that a collision degree corresponding to the acceleration is 0; or if a collision occurs, a speed obtained after the vehicle uses the acceleration may be determined as vvehicle'. In other words, the collision degree R_col is directly used as the initial expectation value Reward. For example, the travelling effect may be obtained through computation according to the following formula (5):

where w1 is a specified fixed coefficient,

represents a speed, after the current acceleration is used, of the vehicle upon a collision, and c is a constant.

[0092] For example, it is assumed that there are a vehicle and a pedestrian. For eight initial accelerations in the action set A, three accelerations at which a collision occurs are determined. In this case, collision degrees R_col1, R_col2, and R_col3 corresponding to the three accelerations may be separately obtained through computation, and a collision degree corresponding to each of the other five accelerations at which no collision occurs is 0.

[0093] After the corresponding travelling effects under the different travelling speed control actions (namely, the different accelerations) are obtained in the foregoing manner, the operation of "selecting a target travelling speed control action from the different travelling speed control actions based on the travelling effects of the vehicle that bring about under the different travelling speed control actions" in S52 may be performed. It may be understood that, an occurrence probability distribution b of intentions is particlized, and a particle set P= {particle1, particle2, ..., particlem} is obtained, representing a mapping relationship b→P related to the occurrence probability distribution b and a particle set with a weight wi. Then, a target expectation value may be determined based on quantities of particles included in various intentions and weights of the particles in the particle set. For example, for the intention g1 of the pedestrian 1, an occurrence probability of the intention g1 may be represented as

, where k meets the following condition: an intention of a particlek

, and wk represents a weight of the particlek.

[0094] For S51, as an example, the operation process may include the following step: predicting N steps based on a mapping relationship: a current initial probability distribution b0→P={particle1, partic/e2,..., particlem}, where the predicted time step is Δt, and it can be obtained that T = NΔt. For the acceleration a, a∈ A={-3, -2, -1, 0, 0.5, 1, 2, 3}, and a travelling effect of the N steps is

, where

Reward (particlek ,a). γ is a discount factor, and is generally a value less than 1. As a quantity N of predicted steps increases, an impact of the discount factor on a decision at a current moment is smaller, and the discount factor is equivalent to a time sequence attenuation factor. It should be noted that an occurrence probability of each intention is specifically reflected by particles of a same intention that are accumulated, and a quantity of particles of a same intention reflects an occurrence probability of the intention. A risk degree of a collision of each intention is specifically reflected in a weight wk of each particle.

[0095] In another example, a collision time of each intention may be further reflected in an interaction probability, an adjusted target intention is determined based on the interaction probability, and a value corresponding to a travelling effect of the vehicle using a target acceleration is computed based on the target intention. Therefore, a target expectation value of the N steps maybe specifically computed according to the following formula:

, where

. It should be noted that an occurrence probability of each intention is specifically reflected in the following: A quantity of particles corresponding to a same intention are accumulated, and the quantity of particles corresponding to the same intention reflects an occurrence probability of the intention. A risk degree of a collision that occurs under each intention is specifically reflected in the following: An interaction probability is computed based on a collision time, a target intention corresponding to each intention is determined, and Reward(particlek, a) is computed based on the target intention.

[0096] In still another example, it may be understood that, to more prominently reflect importance of a risk degree for determining a vehicle speed, and improve reliability and safety of the determined vehicle speed, manners in the foregoing two examples may be further combined to compute, based on each intention, a value corresponding to a travelling effect of the vehicle using a target acceleration. A particle weight that represents particles corresponding to a same intention reflects impact of the risk degree on the vehicle speed, and the interaction probability also reflects impact of the risk degree on the vehicle speed. For example, a value corresponding to a travelling effect of the N steps may be specifically computed according to the following formula:

, where R(bj,a) =

. It should be noted that an occurrence probability of each intention is specifically reflected in the following: A quantity of particles corresponding to a same intention are accumulated, and the quantity of particles corresponding to the same intention reflects an occurrence probability of the intention. A risk degree of a collision that occurs under each intention is specifically reflected in the following: First, a weight wk of each particle corresponding to each intention is used; Second, an interaction probability is computed based on a collision time, a motion status variation corresponding to each intention is determined, and Reward(particlek,a) is computed based on the motion status variation corresponding to each intention.

[0097] It may be understood that, according to the foregoing example computation method, eight initial accelerations in A may be traversed. To be specific, each acceleration is used as a target acceleration, to obtain a travelling effect corresponding to the target acceleration. Finally, eight corresponding travelling effects may be computed, and may be represented as G(b0-3), G(b0-2), G(b0-1), G(b00), G(b00.5), G(b01), G(b02), and G(b03). It should be noted that, the travelling effect is used to indicate a value of a reward function obtained after a target acceleration a is used based on a probability redistribution of various current intentions. A smaller value corresponding to the travelling effect indicates poorer safety. On the contrary, a larger value corresponding to the travelling effect indicates better safety.

[0098] Based on this, it may be understood that, according to a meaning of the value corresponding to the travelling effect, a larger value corresponding to the travelling effect indicates better safety. In this case, in S52, a maximum value corresponding to the travelling effect may be specifically selected from a plurality of values corresponding to the travelling effect, and an acceleration corresponding to the maximum value is determined as the target travelling speed control action, namely, the target acceleration.

[0099] For example, it is assumed that in the determined eight target expectation values: G(b0-3), G(b0-2), G(b0-1), G(b00), G(b00.5), G(b01), G(b02),and G(b03), a maximum value is G(b02). In this case, the corresponding initial acceleration 2 corresponding to G(b02) is selected as the target acceleration. In other words, the target travelling speed control action a=2.

[0100] For S53, in a case, the target acceleration may be directly sent to a controller, and the controller controls the vehicle to travel at the target acceleration. In another case, a target speed of the vehicle may be computed based on the target acceleration and a current speed. For example, the target speed is V = ν0 + a ∗ Δt, where a is the target acceleration, and v0 is the current speed. Then the target speed v is sent to the controller, and the controller controls the vehicle to travel at the target speed v.

[0101] It should be noted that in the foregoing implementation, the Reward value corresponding to the travelling effect is determined based only on the collision degree R col, or the Reward value corresponding to the travelling effect may be determined based on the traffic obstruction degree R_move and/or the travelling discomfort degree R_action. The traffic obstruction degree R move is determined based on a lane speed limit and a vehicle speed reached when the vehicle uses the target acceleration, and the initial expectation value Reward is further determined based on the traffic obstruction degree R_move existing when the vehicle uses the target acceleration. In this case, Reward = R col + R_move, and

, where w2 is a specified fixed coefficient,

is a vehicle speed reached when the vehicle uses a target initial acceleration, and vmax is a speed limit of the current lane. The travelling discomfort degree R action is determined based on the target acceleration and a difference between the target acceleration and the current vehicle speed of the vehicle. The initial expectation value Reward is alternatively determined based on the travelling discomfort degree R_action existing when the vehicle uses the target acceleration. In this case, Reward = R_col + R action, and R action = w3ƒ(actioncurrent) + w4ƒ(actioncurrent - actionlast), where w3 and w4 are specified fixed coefficients, actioncurrent represents a current target acceleration, actionlast represents a target acceleration that is used at a previous moment, f(actioncurrent) represents a comfort return generated when the current target acceleration is used, to suppress travelling discomfort caused by an excessively high acceleration, and f(actioncurrent-actionlast) represents a comfort return generated for a current target acceleration variation, and is used to suppress travelling discomfort caused by an excessively large acceleration variation. It should be noted that the initial expectation reward may alternatively be determined based on the collision degree R col, the traffic obstruction degree R move, and the travelling discomfort degree R action. For each manner of determining the Reward value corresponding to the travelling effect, refer to the foregoing implementation of determining, based only on the collision degree R col, the Reward value corresponding to the travelling effect. Details are not described herein again.

[0102] It should be noted that, in several implementations, only the implementation in which the initial expectation value Reward is determined based on the collision degree R col is used as an example for description. An implementation in which the initial expectation value is determined based on another parameter is similar to this implementation, and details are not described herein again.

[0103] It should be noted that specific implementation of step 503 and step 504 may be specifically implemented by the CPU 205 (the decision-making and planning module 222 in the in-vehicle computer system 220 in FIG. 3 or the pedestrian intention distribution prediction and update module 431 in the decision-making and planning layer 430 in FIG. 4) in FIG. 2. Step 505 may be specifically implemented by the CPU 205 (a speed decision-making and planning unit in the decision-making and planning module 222 in the in-vehicle computer system 220 in FIG. 3 or the speed decision-making and planning module 432 in the decision-making and planning layer 430 in FIG. 4) in FIG. 2.

[0104] It can be learned that in a scenario such as self-driving, according to the method for determining a vehicle speed provided in this embodiment of this application, the probability distribution of intentions may be computed based on the observation information of the surrounding object, and the probability redistribution of the different intentions is computed based on the travel times required for the vehicle to travel from the current position to the risk areas corresponding to the different intentions. Further, the motion status variations of the surrounding object with the different intentions are predicted based on the travel times required for the vehicle to travel to the risk areas corresponding to the different intentions. Finally, the travelling speed of the vehicle is determined based on the probability redistribution of the different intentions, the motion status variations of the surrounding object with the different intentions, and the motion status variations of the vehicle under different travelling speed control actions.

[0105] Therefore, in a travelling process of the vehicle, for a plurality of intentions of the surrounding object of the vehicle, a probability of each intention may be predicted based on the observation information of the surrounding object. In addition, a risk degree of a collision, between the surrounding object and the vehicle, that is corresponding to each intention and that is under control of each acceleration of the vehicle is predicted. Then, the vehicle speed is determined based on the probability of each intention and the risk degree of a collision. In this way, during determining of the vehicle speed, each possible intention of the surrounding object is considered, and the risk degree of a collision, between the surrounding object and the vehicle, that is corresponding to each intention and that is under control of each acceleration of the vehicle is further considered. In this way, a high risk, between the surrounding object and the vehicle, that is less likely to occur is not ignored. Therefore, a determined travelling speed is more appropriate for a current driving environment, and a potential safety risk during travelling of the vehicle is reduced.

[0106] In addition, an embodiment of this application further provides an apparatus for determining a vehicle speed. Referring to FIG. 11, the apparatus 1100 includes a first obtaining unit 1101, a first computation unit 1102, a second computation unit 1103, a prediction unit 1104, and a first determining unit 1105.

[0107] The first obtaining unit 1101 is configured to obtain observation information of a surrounding object of a vehicle. The first computation unit 1102 is configured to compute, based on the observation information of the surrounding object, a probability distribution of different intentions of the surrounding object.

[0108] The second computation unit 1103 is configured to perform redistribution computation on the probability distribution based on travel times required for the vehicle to travel from a current position of the vehicle to risk areas corresponding to the different intentions, to obtain a probability redistribution of the different intentions, where the risk areas corresponding to the different intentions are areas, in a lane in which the vehicle travels, located proximate to the surrounding object with the different intentions.

[0109] The prediction unit 1104 is configured to predict motion status variations of the surrounding object with the different intentions based on the travel times required for the vehicle to travel to the risk areas corresponding to the different intentions.

[0110] The first determining unit 1105 is configured to determine a travelling speed of the vehicle based on the probability redistribution of the different intentions, the motion status variations of the surrounding object with the different intentions, and motion status variations of the vehicle under different travelling speed control actions.

[0111] In a possible implementation, the first computation unit 1102 may include an establishment subunit and a computation subunit. The establishment subunit is configured to establish, based on the observation information of the surrounding object by establishing a coordinate system related to the lane in which the vehicle travels, a relative position relationship between the surrounding object and the lane and a relative motion relationship between the surrounding object and the lane. The computation subunit is configured to compute the probability distribution of the different intentions of the surrounding object based on the relative position relationship between the surrounding object and the lane and the relative motion relationship between the surrounding object and the lane.

[0112] In another possible implementation, the apparatus may further include a second obtaining unit, an establishment unit, a second determining unit, and a third computation unit.

[0113] The second obtaining unit is configured to obtain observation information of the vehicle. The establishment unit is configured to establish, based on the observation information of the vehicle and the observation information of the surrounding object by establishing a coordinate system related to the lane in which the vehicle travels, a relative position relationship between the vehicle and the lane and a relative motion status between the vehicle and the lane, and a relative position relationship between the surrounding object and the lane and a relative motion status between the surrounding object and the lane. The second determining unit is configured to determine, based on the relative position relationship between the surrounding object and the lane and the relative motion status between the surrounding object and the lane, the risk areas corresponding to the different intentions. The third computation unit is configured to compute, based on the relative position relationship between the vehicle and the lane and the relative motion status between the vehicle and the lane, and the risk areas corresponding to the different intentions, the travel times required for the vehicle to travel from the current position of the vehicle to the risk areas corresponding to the different intentions.

[0114] In still another possible implementation, the second computation unit 1103 may include a processing subunit and an adjustment subunit. The processing subunit is configured to perform particle processing on the probability distribution, where quantities of particles corresponding to the different intentions are used to represent the probability distribution of the different intentions. The adjustment subunit is configured to adjust, based on the travel times that are obtained through computation and that are required for the vehicle to travel to the risk areas corresponding to the different intentions, weights of the particles corresponding to the different intentions, to obtain the probability redistribution of the different intentions.

[0115] In still yet another possible implementation, the prediction unit 1104 may include a first determining subunit and a prediction subunit. The first determining subunit is configured to determine, based on the travel times required for the vehicle to travel to the risk areas corresponding to the different intentions, probabilities that the surrounding object with the different intentions changes the current intention. The prediction subunit is configured to predict the motion status variations of the surrounding object with the different intentions based on the probabilities that the surrounding object with the different intentions changes the current intention and a random probability.

[0116] In a further possible implementation, the first determining unit 1105 may include an estimation subunit, a selection subunit, and a second determining subunit. The estimation subunit is configured to estimate travelling effects of the vehicle that bring about under the different travelling speed control actions based on the probability redistribution of the different intentions, the motion status variations of the surrounding object with the different intentions, and the motion status variations of the vehicle under the different travelling speed control actions. The selection subunit is configured to select a target travelling speed control action from the different travelling speed control actions based on the travelling effects of the vehicle that bring about under the different travelling speed control actions. The second determining subunit is configured to determine the travelling speed of the vehicle based on the target travelling speed control action.

[0117] It should be noted that the apparatus 1100 is configured to perform the steps in the embodiment corresponding to FIG. 5, To be specific, the obtaining unit 1101 may specifically perform step 501; the first computation unit 1102 may specifically perform step 502; the second computation unit 1103 may specifically perform step 503; the prediction unit 1104 may specifically perform step 504; and the first determining unit 1105 may specifically perform step 505. It may be understood that the apparatus 1100 is corresponding to the method for determining a vehicle speed provided in the embodiments of this application. Therefore, for implementations of the apparatus 1100 and technical effects that can be achieved by the apparatus 1100, refer to the related descriptions of the implementations of the method for determining a vehicle speed in the embodiments of this application.

[0118] In addition, an embodiment of this application further provides a vehicle. Referring to FIG. 12, the vehicle 1200 includes a sensor 1201, a processor 1202, and a vehicle speed controller 1203.

[0119] The sensor 1201 is configured to: obtain observation information of a surrounding object of the vehicle, and send the observation information to the processor, for example, a radar or a camera.

[0120] The processor 1202 is configured to: determine a travelling speed of the vehicle according to the method in any one of the implementations of the first aspect, and send the travelling speed to the vehicle speed controller.

[0121] The vehicle speed controller 1203 is configured to control the vehicle to travel at the determined travelling speed of the vehicle.

[0122] It may be understood that the vehicle 1200 performs the method for determining a vehicle speed provided in the embodiments of this application. Therefore, for implementations of the vehicle 1200 and technical effects that can be achieved by the vehicle 1200, refer to the related descriptions of the implementations of the method for determining a vehicle speed in the embodiments of this application.

[0123] In addition, an embodiment of this application further provides a vehicle. Referring to FIG. 13, the vehicle 1300 includes a processor 1301 and a memory 1302. The memory 1302 stores an instruction, and when the processor 1301 executes the instruction, the vehicle 1300 is enabled to perform the method in any one of the implementations of the method for determining a vehicle speed.

[0124] It may be understood that the vehicle 1300 performs the method for determining a vehicle speed provided in the embodiments of this application. Therefore, for implementations of the vehicle 1300 and technical effects that can be achieved by the vehicle 1300, refer to the related descriptions of the implementations of the method for determining a vehicle speed in the embodiments of this application.

[0125] In addition, an embodiment of this application further provides a computer program product. When the computer program product runs on a computer, the computer performs the method in any one of the implementations of the method for determining a vehicle speed.

[0126] In addition, an embodiment of this application further provides a computer-readable storage medium. The computer-readable storage medium stores an instruction. When the instruction is run on a computer or a processor, the computer or the processor is enabled to perform the method in any one of the implementations of the method for determining a vehicle speed.

[0127] "First" in terms such as "first risk degree" mentioned in the embodiments of this application is used only for name identification, and does not indicate the first in sequence. This rule is also applicable to "second" and the like.

[0128] From the foregoing descriptions of the implementations, a person skilled in the art may clearly understand that all or some steps of the method in the foregoing embodiment may be implemented by using a combination of software and a universal hardware platform. Based on such an understanding, the technical solutions of this application may be implemented in a form of a software product. The software product may be stored in a storage medium, for example, a read-only memory (English: read-only memory, ROM)/RAM, a magnetic disk, or an optical disc, and includes several instructions for instructing a computer device (which may be a personal computer, a server, or a network communications device such as a router) to perform the method described in the embodiments or some parts of the embodiments of this application.

[0129] The embodiments in this specification are all described in a progressive manner. For same or similar parts in the embodiments, reference may be made to these embodiments, and each embodiment focuses on a difference from other embodiments. Especially, an apparatus embodiment is basically similar to a method embodiment, and therefore is described briefly. For related parts, reference may be made to some descriptions in the method embodiment. The described apparatus embodiment is merely an example. The modules described as separate parts may or may not be physically separate, and parts displayed as modules may or may not be physical modules, may be located in one position, or may be distributed on a plurality of network units. Some or all of the modules may be selected based on actual needs to achieve the objectives of the solutions of the embodiments. A person of ordinary skill in the art may understand and implement the embodiments of the present invention without creative efforts.

[0130] The foregoing descriptions are merely example implementations of this application, but are not intended to limit the protection scope of this application.


Claims

1. A method for determining a vehicle speed, comprising:

obtaining observation information of a surrounding object of a vehicle;

computing, based on the observation information of the surrounding object, a probability distribution of different intentions of the surrounding object;

performing redistribution computation on the probability distribution based on travel times required for the vehicle to travel from a current position of the vehicle to risk areas corresponding to the different intentions, to obtain a probability redistribution of the different intentions, wherein the risk areas corresponding to the different intentions are areas, in a lane in which the vehicle travels, located proximate to the surrounding object with the different intentions;

predicting motion status variations of the surrounding object with the different intentions based on the travel times required for the vehicle to travel to the risk areas corresponding to the different intentions; and

determining a travelling speed of the vehicle based on the probability redistribution of the different intentions, the motion status variations of the surrounding object with the different intentions, and motion status variations of the vehicle under different travelling speed control actions.


 
2. The method according to claim 1, wherein the computing, based on the observation information of the surrounding object, a probability distribution of different intentions of the surrounding object comprises:

establishing, based on the observation information of the surrounding object by establishing a coordinate system related to the lane in which the vehicle travels, a relative position relationship between the surrounding object and the lane and a relative motion relationship between the surrounding object and the lane; and

computing the probability distribution of the different intentions of the surrounding object based on the relative position relationship between the surrounding object and the lane and the relative motion relationship between the surrounding object and the lane.


 
3. The method according to claim 1, further comprising:

obtaining observation information of the vehicle;

establishing, based on the observation information of the vehicle and the observation information of the surrounding object by establishing a coordinate system related to the lane in which the vehicle travels, a relative position relationship between the vehicle and the lane and a relative motion status between the vehicle and the lane, and a relative position relationship between the surrounding object and the lane and a relative motion status between the surrounding object and the lane;

determining, based on the relative position relationship between the surrounding object and the lane and the relative motion status between the surrounding object and the lane, the risk areas corresponding to the different intentions; and

computing, based on the relative position relationship between the vehicle and the lane and the relative motion status between the vehicle and the lane, and the risk areas corresponding to the different intentions, the travel times required for the vehicle to travel from the current position of the vehicle to the risk areas corresponding to the different intentions.


 
4. The method according to any one of claims 1 to 3, wherein the performing redistribution computation on the probability distribution based on travel times required for the vehicle to travel to risk areas corresponding to the different intentions, to obtain a probability redistribution of the different intentions comprises:

performing particle processing on the probability distribution, wherein quantities of particles corresponding to the different intentions are used to represent the probability distribution of the different intentions; and

adjusting, based on the travel times that are obtained through computation and that are required for the vehicle to travel to the risk areas corresponding to the different intentions, weights of the particles corresponding to the different intentions, to obtain the probability redistribution of the different intentions.


 
5. The method according to any one of claims 1 to 4, wherein the predicting motion status variations of the surrounding object with the different intentions based on the travel times required for the vehicle to travel to the risk areas corresponding to the different intentions comprises:

determining, based on the travel times required for the vehicle to travel to the risk areas corresponding to the different intentions, probabilities that the surrounding object with the different intentions changes the current intention; and

predicting the motion status variations of the surrounding object with the different intentions based on the probabilities that the surrounding object with the different intentions changes the current intention and a random probability.


 
6. The method according to any one of claims 1 to 5, wherein the determining a travelling speed of the vehicle based on the probability redistribution of the different intentions, the motion status variations of the surrounding object with the different intentions, and motion status variations of the vehicle under different travelling speed control actions comprises:

estimating travelling effects of the vehicle that bring about under the different travelling speed control actions based on the probability redistribution of the different intentions, the motion status variations of the surrounding object with the different intentions, and the motion status variations of the vehicle under the different travelling speed control actions;

selecting a target travelling speed control action from the different travelling speed control actions based on the travelling effects of the vehicle that bring about under the different travelling speed control actions; and

determining the travelling speed of the vehicle based on the target travelling speed control action.


 
7. An apparatus for determining a vehicle speed, comprising:

a first obtaining unit, configured to obtain observation information of a surrounding object of a vehicle;

a first computation unit, configured to compute, based on the observation information of the surrounding object, a probability distribution of different intentions of the surrounding object;

a second computation unit, configured to perform redistribution computation on the probability distribution based on travel times required for the vehicle to travel from a current position of the vehicle to risk areas corresponding to the different intentions, to obtain a probability redistribution of the different intentions, wherein the risk areas corresponding to the different intentions are areas, in a lane in which the vehicle travels, located proximate to the surrounding object with the different intentions;

a prediction unit, configured to predict motion status variations of the surrounding object with the different intentions based on the travel times required for the vehicle to travel to the risk areas corresponding to the different intentions; and

a first determining unit, configured to determine a travelling speed of the vehicle based on the probability redistribution of the different intentions, the motion status variations of the surrounding object with the different intentions, and motion status variations of the vehicle under different travelling speed control actions.


 
8. The apparatus according to claim 7, wherein the first computation unit comprises:

an establishment subunit, configured to establish, based on the observation information of the surrounding object by establishing a coordinate system related to the lane in which the vehicle travels, a relative position relationship between the surrounding object and the lane and a relative motion relationship between the surrounding object and the lane; and

a computation subunit, configured to compute the probability distribution of the different intentions of the surrounding object based on the relative position relationship between the surrounding object and the lane and the relative motion relationship between the surrounding object and the lane.


 
9. The apparatus according to claim 7, wherein the apparatus further comprises:

a second obtaining unit, configured to obtain observation information of the vehicle;

an establishment unit, configured to establish, based on the observation information of the vehicle and the observation information of the surrounding object by establishing a coordinate system related to the lane in which the vehicle travels, a relative position relationship between the vehicle and the lane and a relative motion status between the vehicle and the lane, and a relative position relationship between the surrounding object and the lane and a relative motion status between the surrounding object and the lane;

a second determining unit, configured to determine, based on the relative position relationship between the surrounding object and the lane and the relative motion status between the surrounding object and the lane, the risk areas corresponding to the different intentions; and

a third computation unit, configured to compute, based on the relative position relationship between the vehicle and the lane and the relative motion status between the vehicle and the lane, and the risk areas corresponding to the different intentions, the travel times required for the vehicle to travel from the current position of the vehicle to the risk areas corresponding to the different intentions.


 
10. The apparatus according to any one of claims 7 to 9, wherein the second computation unit comprises:

a processing subunit, configured to perform particle processing on the probability distribution, wherein quantities of particles corresponding to the different intentions represent the probability distribution of the different intentions; and

an adjustment subunit, configured to adjust, based on the travel times that are obtained through computation and that are required for the vehicle to travel to the risk areas corresponding to the different intentions, weights of the particles corresponding to the different intentions, to obtain the probability redistribution of the different intentions.


 
11. The apparatus according to any one of claims 7 to 10, wherein the prediction unit comprises:

a first determining subunit, configured to determine, based on the travel times required for the vehicle to travel to the risk areas corresponding to the different intentions, probabilities that the surrounding object with the different intentions changes the current intention; and

a prediction subunit, configured to predict the motion status variations of the surrounding object with the different intentions based on the probabilities that the surrounding object with the different intentions changes the current intention and a random probability.


 
12. The apparatus according to any one of claims 7 to 11, wherein the first determining unit comprises:

an estimation subunit, configured to estimate travelling effects of the vehicle that bring about under the different travelling speed control actions based on the probability redistribution of the different intentions, the motion status variations of the surrounding object with the different intentions, and the motion status variations of the vehicle under the different travelling speed control actions;

a selection subunit, configured to select a target travelling speed control action from the different travelling speed control actions based on the travelling effects of the vehicle that bring about under the different travelling speed control actions; and

a second determining subunit, configured to determine the travelling speed of the vehicle based on the target travelling speed control action.


 




Drawing





































Search report













Cited references

REFERENCES CITED IN THE DESCRIPTION



This list of references cited by the applicant is for the reader's convenience only. It does not form part of the European patent document. Even though great care has been taken in compiling the references, errors or omissions cannot be excluded and the EPO disclaims all liability in this regard.

Patent documents cited in the description