(19)
(11)EP 3 535 096 B1

(12)EUROPEAN PATENT SPECIFICATION

(45)Mention of the grant of the patent:
02.09.2020 Bulletin 2020/36

(21)Application number: 17794817.1

(22)Date of filing:  11.10.2017
(51)International Patent Classification (IPC): 
B25J 9/16(2006.01)
G05D 1/00(2006.01)
G05B 19/401(2006.01)
G01B 11/24(2006.01)
(86)International application number:
PCT/US2017/056072
(87)International publication number:
WO 2018/085013 (11.05.2018 Gazette  2018/19)

(54)

ROBOTIC SENSING APPARATUS AND METHODS OF SENSOR PLANNING

ROBOTISCHE MESSVORRICHTUNG UND VERFAHREN ZUR SENSORPLANUNG

APPAREIL DE DÉTECTION ROBOTIQUE ET PROCÉDÉS DE PLANIFICATION DE CAPTEURS


(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

(30)Priority: 03.11.2016 US 201615342500

(43)Date of publication of application:
11.09.2019 Bulletin 2019/37

(73)Proprietor: General Electric Company
Schenectady, NY 12345 (US)

(72)Inventors:
  • LIM, Ser, Nam
    Niskayuna NY 12309 (US)
  • DIWINSKY, David, Scott
    Cincinnati OH 45215 (US)
  • BIAN, Xiao
    Niskayuna NY 12309-1027 (US)
  • GRADY, Wayne, Ray
    Cincinnati OH 45215-1988 (US)
  • UZUNBAS, Mustafa, Gokhan
    Niskayuna NY 12309-1027 (US)
  • KABA, Mustafa, Devrim
    Niskayuna NY 12309 (US)

(74)Representative: Openshaw & Co. 
8 Castle Street
Farnham, Surrey GU9 7HR
Farnham, Surrey GU9 7HR (GB)


(56)References cited: : 
US-A1- 2002 169 586
US-A1- 2015 254 831
US-A1- 2005 284 937
US-A1- 2016 046 373
  
  • WILLIAM R SCOTT: "Model-based view planning", MACHINE VISION AND APPLICATIONS, SPRINGER, BERLIN, DE, vol. 20, no. 1, 30 November 2007 (2007-11-30), pages 47-69, XP019651723, ISSN: 1432-1769
  • PALETTA L ET AL: "Active object recognition by view integration and reinforcement learning", ROBOTICS AND AUTONOMOUS SYST, ELSEVIER SCIENCE PUBLISHERS, AMSTERDAM, NL, vol. 31, no. 1-2, 30 April 2000 (2000-04-30), pages 71-86, XP004218291, ISSN: 0921-8890, DOI: 10.1016/S0921-8890(99)00079-2
  
Note: Within nine months from the publication of the mention of the grant of the European patent, any person may give notice to the European Patent Office of opposition to the European patent granted. Notice of opposition shall be filed in a written reasoned statement. It shall not be deemed to have been filed until the opposition fee has been paid. (Art. 99(1) European Patent Convention).


Description

FIELD



[0001] The present disclosure is related generally to robotic sensing devices and methods of sensor planning.

BACKGROUND



[0002] Sensor planning is a general requirement in inspections, measurements, and robot localization, navigation, or mapping relative to an area of interest. Areas of interest may include a component, part, detail, assembly, or spatial area, such as a geographic area or other 2D or 3D space. Additionally, a general requirement of inspection and measurement methods, and autonomous robotics, is to employ sensors to capture samples, such as images or measurements, of the area of interest in sufficient detail and a desired level of completeness.

[0003] A known solution for sensor planning is to utilize manually programmed sensor plans, such as coordinate systems, routes, or pathways, for capturing the desired area of interest. However, manually programmed sensor plans often require unchanging and/or substantially certain areas of interest. Therefore, areas of interest that deviate from the preprogrammed plan often result in sampling errors, omissions, or other failures.

[0004] Another known solution for sensor planning may include programming a robot to capture large quantities of samples to ensure the area of interest is captured in sufficient detail and a desired level of completeness. However, capturing large quantities of samples is similarly costly, time consuming, and results in inefficient quantities of redundant samples. Additionally, the unknown nominal areas of interest, or changes to the area of interest relative to nominal, may similarly result in errors, omissions, or failures to capture the desired area of interest.

[0005] Therefore, there exists a need for robotic sensing systems and methods of sensor planning that may capture samples of the desired and/or potentially unknown or changing area of interest in sufficient detail and completeness while minimizing redundancy and time.

[0006] W R SCOTT, "Model-based view planning", Machine Vision and Applications, Springer, Berlin (DE), vol. 20, no. 1, pages 47-69 (30 November 2007) [XP019651723] describes a multi-phase, model-based approach to view planning for automated, high fidelity object inspection or reconstruction by means of laser scanning range sensors.

BRIEF DESCRIPTION



[0007] The present disclosure is directed to a computer-implemented method of sensor planning for acquiring samples via an apparatus including one or more sensors. The computer-implemented method includes defining, by one or more computing devices, an area of interest; identifying, by the one or more computing devices, one or more sensing parameters for the one or more sensors; determining, by the one or more computing devices, one or more sampling combinations for acquiring a plurality of samples of the area of interest by the one or more sensors based at least in part on the one or more sensing parameters, comprising determining a combination of overlap exponentials based at least on a reinforcement learning algorithm, wherein an overlap exponential is a factor by which overlap between the sample and a previous sample is encouraged, calculating a score function for one or more sampling combinations based at least on a total area covered by the one or more sensors, an overlap perimeter, and the combination of overlap exponentials, wherein an overlap perimeter is a quantity of the sample that is redundant relative to a previous sample, and selecting, the sampling combination corresponding to a maximum score function; and providing, by the one or more computing devices, one or more command control signals to the apparatus including the one or more sensors to acquire the plurality of samples of the area of interest using the one or more sensors based at least on the sampling combination.

[0008] A further aspect of the present disclosure is directed to a robotic sensing apparatus for sensor planning. The apparatus includes one or more sensors and a computing device, in which the computing device includes one or more processors and one or more memory devices. The one or more memory devices store instructions that when executed by the one or more processors cause the one or more processors to perform operations. The operations include receiving an area of interest; receiving one or more sensing parameters for the one or more sensors; determining one or more sampling combinations for acquiring a plurality of samples of the area of interest by the one or more sensors, comprising determining a combination of overlap exponentials based at least on a reinforcement learning algorithm, wherein an overlap exponential is a factor by which overlap between the sample and a previous sample is encouraged, calculating a score function for one or more sampling combinations based at least on a total area covered by the one or more sensors, an overlap perimeter, and the combination of overlap exponentials, wherein an overlap perimeter is a quantity of the sample that is redundant relative to a previous sample, and selecting, the sampling combination corresponding to a maximum score function; and acquiring the plurality of samples of the area of interest using the one or more sensors based at least on the sampling combination.

[0009] These and other features, aspects and advantages of the present invention will become better understood with reference to the following description and appended claims. The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments of the invention and, together with the description, serve to explain the principles of the invention.

BRIEF DESCRIPTION OF THE DRAWINGS



[0010] A full and enabling disclosure of the present invention, including the best mode thereof, directed to one of ordinary skill in the art, is set forth in the specification, which makes reference to the appended figures, in which:

FIG. 1 is an exemplary embodiment of a robotic sensing apparatus;

FIG. 2 is an exemplary embodiment of another robotic sensing apparatus;

FIG. 3 is a flowchart outlining an exemplary method of sensor planning; and

FIG. 4 is an exemplary embodiment of yet another robotic sensing apparatus.



[0011] Repeat use of reference characters in the present specification and drawings is intended to represent the same or analogous features or elements of the present invention.

DETAILED DESCRIPTION



[0012] Reference now will be made in detail to embodiments of the invention, one or more examples of which are illustrated in the drawings. Each example is provided by way of explanation of the invention, not limitation of the invention. In fact, it will be apparent to those skilled in the art that various modifications and variations can be made in the present invention without departing from the scope of the invention as defined in the appended claims. For instance, features illustrated or described as part of one embodiment can be used with another embodiment to yield a still further embodiment. Thus, it is intended that the present invention covers such modifications and variations as come within the scope of the appended claims.

[0013] As used herein, the terms "first", "second", and "third" may be used interchangeably to distinguish one component from another and are not intended to signify location or importance of the individual components.

[0014] Robotic sensing apparatuses and methods of sensor planning that may capture samples of an area of interest while minimizing redundancy and time are generally provided. The methods and systems described herein may include steps or operations that may capture samples of an area of interest, such as a component, or assembly, or geographic area, at a desired resolution and level of completeness while minimizing the quantity of samples, such as images or measurements, taken to capture the area of interest. Various embodiments of the robotic sensing apparatuses and methods described herein may utilize a deep learning approach in conjunction with sensor planning. Furthermore, the systems and methods described herein may generally autonomously plan and capture a minimal quantity of samples to capture the area of interest at a desired level of completeness.

[0015] Referring now to FIGS. 1 and 2, a robotic sensing apparatus for sensor planning 90 (herein referred to as "apparatus 90") includes one or more sensors 110 acquiring samples of an area of interest 130. The one or more sensors 110 may include an imaging device, a proximity sensor, or combinations thereof. In one embodiment, imaging devices may generally include cameras. In another embodiment, imaging devices may specifically include interferometers, such as, but not limited to, optical coherence tomography (e.g., white light scanners or blue light scanners). In other embodiments, the one or more sensors 110 include proximity sensors, in which the proximity sensors may generally include sensors that may emit and/or retrieve electromagnetic signals and process changes in said electromagnetic signals. For example, proximity sensors may include, but are not limited to, capacitive, infrared, inductive, magnetic, sonic or ultrasonic proximity sensors, radar, LIDAR, or laser rangefinders. In various embodiments, the one or more sensors 110 may include combinations of imaging devices and/or proximity sensors. In various embodiments, the one or more sensors 110 acquire samples, including images or measurements, at various resolutions, angles, distances, orientations, sampling or measurement rates, frequencies, etc.

[0016] In one embodiment, the apparatus 90 includes a translatable robotic apparatus 100 (herein referred to as "robot 100"). The robot 100 may include a movable fixture, such as a robotic arm as shown in FIGS. 1 and 2, or an autonomous mobile vehicle, such as a drone as shown in FIG. 4. Translations of the robot 100, the one or more sensors 110, and/or the area of interest 130 may include, but are not limited to, six-axis movements (e.g. up/down, side/side, forward/backward, etc.), pivots, turns, rotations, and/or displacements at constant or variable rates of motion. In the embodiments shown in FIGS. 1 and 2, the sensor(s) 110 may be mounted to the robot 100 in which the robot 100 translates the sensor 110 to various portions 131 of an area of interest 130 at various angles or distances relative to the area of interest 130.

[0017] In other embodiments of the apparatus 90, the robot 100 may translate the area of interest 130 relative to the one or more sensors 110. For example, the robot 100, such as a robotic arm, may translate the area of interest 130 relative to one or more fixed sensors 110. The robot 100 may translate the area of interest 130 to various distances, angles, and/or orientations relative to the one or more sensors 110.

[0018] Referring now to FIG. 3, a flowchart outlining steps of an exemplary embodiment of a method of sensor planning 300 (herein referred to as "method 300") is generally provided. The method 300 shown in FIG. 3 may be implemented by the apparatus 90 shown and described in regard to FIGS. 1 and 2. The method 300 may further be implemented by one or more computing devices, such as the computing device 120 described in regard to FIG. 4. FIG. 3 depicts steps performed in a particular order for purposes of illustration and discussion. Those of ordinary skill in the art, using the disclosures provided herein, will understand that various steps of any of the methods disclosed herein can be modified, adapted, expanded, rearranged and/or omitted in various ways without deviating from the scope of the present disclosure.

[0019] The method 300 can include at (310) defining, by one or more computing devices, an area of interest, at (320) identifying, by the one or more computing devices, one or more sensing parameters for the one or more sensors, at (330) determining, by the one or more computing devices, a sampling combination for acquiring a plurality of samples by the one or more sensors based at least in part on the one or more sensing parameters, and at (340) providing, by one or more computing devices, one or more command control signals to an apparatus including the one or more sensors to acquire the plurality of samples of the area of interest using the one or more sensors based at least on the sampling combination.

[0020] At (310), the method 300 may include defining an area of interest. In one embodiment, defining an area of interest includes receiving a point cloud. Receiving a point cloud may include receiving an image file, such as a computer-aided design (CAD) file, of the area of interest. The image file may include a nominal file of the area of interest to which the samples from the sensors may measure in comparison.

[0021] In another embodiment, defining an area of interest may include defining a finite space in which a robot and/or one or more sensors may operate, such as the robot 100 and/or one or more sensors 110 shown in FIGS. 1, 2, and 4. For example, the extent to which the robot 100 may translate may be spatially limited. In one example, the robot 100, as a robotic arm, may be limited in its range of motion, extension, etc. In another example, the robot 100, as a drone, may be geographically limited by coordinates, operating range, or operating envelope, such as altitude, speed, maneuverability, etc. Therefore, defining an area of interest may include defining a 2D or 3D space in which samples may be taken.

[0022] In still other embodiments, defining an area of interest may include taking a sample of the area of interest. For example, taking a sample of the area of interest may include taking a sample that broadly captures the area of interest, including a perimeter of the area of interest. Broadly capturing the area of interest may include sampling at a low resolution, or a large distance from the area of interest, or otherwise in minimal detail to obtain and define a periphery of the area of interest. In various embodiments, broadly capturing the area of interest may include capturing the defined spatial area as limited by coordinates, operating range, operating envelope, etc. In other embodiments, broadly capturing the area of interest may be dependent on a maximum sampling area of one or more sensors such that the area of interest may be defined by the one or more sensing parameters.

[0023] At (320), the method 300 includes identifying one or more sensing parameters for the one or more sensors. In one embodiment, identifying one or more sensing parameters may include defining one or more of a measurement resolution, a field of view, and/or a depth of field. Defining the measurement resolution may include defining a lateral resolution, a lens resolution, an image resolution, and/or a sensor resolution. Defining a sensor resolution may include defining a spatial and/or temporal sensor resolution.

[0024] In another embodiment, identifying one or more sensing parameters may include defining a total area covered by the one or more sensors. For example, defining a total area covered by the one or more sensors may be a function of one or more of the defined aforementioned resolutions. As another non-limiting example, such as shown and described in regard to FIGS. 1, 2, and 4, defining a total area covered may be approximately equal to or less than the portion 131 of the area of interest 130 captured by the one or more sensors 110. In one embodiment, the total area covered by the one or more sensors may be a function of one or more of the defined aforementioned resolutions and additional user-defined limits based at least on a desired sample quality. For example, defining the total area covered may include defined hardware capabilities of the one or more sensors and a subset of said hardware capabilities based at least on a user-defined limitation. The user-defined limitation may be based generally on good visibility standards as defined by the user. Good visibility standards may be based at least on a measurement resolution, a field of view, and/or depth of field of the one or more sensors.

[0025] In still other embodiments at (320), identifying one or more sensing parameters may include calculating a curvature and/or normal vector of at least a portion of the area of interest. Calculating the curvature and/or normal vector may be based at least on a surface of the point cloud or image file defined in (310). The normal vector may define one or more sensor centers. For example, referring to FIGS. 1 and 2, the normal vector may define one or more centerlines 111 of the one or more sensors 110 relative to the portion 131 of the area of interest 130. As another non-limiting example, the normal vector may define an angle of view 112 of the sensor 110 relative to the area of interest 130, or the portion 131 thereof.

[0026] At (330), the method 300 includes determining a sampling combination for acquiring a plurality of samples by the one or more sensors based at least in part on the one or more sensing parameters. The sampling combination may be a combination of samples of the area of interest pursuant to capturing the area of interest. The combination of samples of the area of interest may include translations of the sensor(s) and/or the area of interest relative to one another. The sampling combination may further be a combination of sensing parameters relative to translations of the sensor(s) and/or the area of interest. In various embodiments, the sampling combination may include combinations of samples of various portions of the area of interest taken to capture the area of interest. For example, referring to FIGS. 1, 2 or 4, the sampling combination may include a specific sequence of translations and/or sensing parameters at various portions 131 of the area of interest 130 until the area of interest 130 is captured. The specific sequence of translations and/or sensing parameters may include distances, angles, and/or resolutions of the one or more sensor 110 relative to the area of interest 130 for each sample captured of the portion 131 of the area of interest 130.

[0027] Determining a sampling combination to be acquired by the one or more sensors may further include determining a minimal quantity of samples to acquire to capture the area of interest. In one embodiment, determining a sampling combination to be acquired by the one or more sensors may include selecting the sampling combination based at least on a score function,

for one or more sampling combinations (c0, c1, ..., cr). The total area covered is based at least on one or more sensing parameters or the portion of the area of interest. The overlap perimeter is a quantity of the sample that is redundant (e.g. overlapping) relative to a previous sample. Lambda λ is an overlap exponential. The overlap exponential is a factor by which overlap between the sample and a previous sample is encouraged. For example, λ = 0 may discourage overlap and encourage sampling combinations including samples with large total areas covered. However, λ = 0 may result in sampling combinations in which portions of the area of interest are uncaptured between the samples. As another example, λ > 0 may encourage overlap to ensure portions of the area of interest between the samples (e.g. gaps) are captured. However, λ > 0 may result in large quantities of samples, or translations of the sensor(s) or area of interest, for a given sampling combination to capture the area of interest.

[0028] In another embodiment, determining a sampling combination to be acquired by the one or more sensors may include at (332) determining a combination of overlap exponentials based at least on a reinforcement learning (RL) algorithm, at (334) calculating a score function for one or more sampling combinations based at least on a total area covered by the one or more sensors, an overlap perimeter, and the one or more overlap exponentials, and at (336) selecting the sampling combination corresponding to a maximum score function.

[0029] At (332), the method 300 may include using at least one of a State-Action-Result-State-Action (SARSA), Q-Learning, and Policy Gradient RL algorithm to determine a combination of overlap exponentials that may output a sampling combination that may minimize a quantity of samples taken of the area of interest pursuant to capturing the area of interest to a desired level of completeness. In various embodiments, at least one of the Q-Learning and Policy Gradient PL algorithms may be used in conjunction with a deep learning approach to determine a minimal quantity of samples for capturing the area of interest at a desired level of completeness. Determining the combination of overlap exponentials may include determining combinations of zero and non-zero overlap exponentials that may result in a maximum score function while capturing the area of interest in a minimal quantity of samples to a desired level of completeness.

[0030] At (340), the method 300 includes acquiring the plurality of samples using the one or more sensors based at least on the sampling combination. In one embodiment, acquiring the plurality of samples may include translating the one or more sensor(s) and/or the area of interest relative to one another. For example, referring to FIGS. 1, 2, or 4, the one or more sensors 110 and/or the area of interest 130 may be mounted to the robot 100 and translate to capture samples at a plurality of portions 131 of the area of interest 130 until the area of interest 130 is captured in desired detail and completeness. In another embodiment, the method 300 may be implemented to determine positions, placements, setups, orientations, distances, etc. of one or more sensors relative to the area of interest using the determined sampling combination. For example, within the defined area of interest, such as a 2D or 3D space, the determined sampling combination may provide positions, placements, and orientations of sensors such that a minimal quantity of sensors is utilized to capture the area of interest within the 2D or 3D space to a desired level of detail and completeness.

[0031] FIG. 4 depicts an example apparatus 90 according to exemplary embodiments of the present disclosure. The apparatus 90 can include one or more sensors 110, a robot 100, and one or more computing devices 120. In one embodiment, the robot 100 defines a robotic arm such as shown in regard to FIGS. 1 and 2. In another embodiment, such as shown in FIG. 4, the robot 100 defines an autonomous mobile vehicle, such as a drone. The one or more sensors 120, the robot 100, and/or the computing device 120 can be configured to communicate via more or more network(s) 410, which can include any suitable wired and/or wireless communication links for transmission of the communications and/or data, as described herein. For instance, the network 410 can include a SATCOM network, ACARS network, ARINC network, SITA network, AVICOM network, a VHF network, a HF network, a Wi-Fi network, a WiMAX network, a gatelink network, etc.

[0032] The computing device(s) 120 can include one or more processor(s) 121 and one or more memory device(s) 122. The one or more processor(s) 121 can include any suitable processing device, such as a microprocessor, microcontroller, integrated circuit, logic device, and/or other suitable processing device. The one or more memory device(s) 122 can include one or more computer-readable media, including, but not limited to, non-transitory computer-readable media, RAM, ROM, hard drives, flash drives, and/or other memory devices.

[0033] The one or more memory device(s) 122 can store information accessible by the one or more processor(s) 121, including computer-readable instructions 123 that can be executed by the one or more processor(s) 121. The instructions 123 can be any set of instructions that when executed by the one or more processor(s) 121, cause the one or more processor(s) 121 to perform operations. In some embodiments, the instructions 123 can be executed by the one or more processor(s) 121 to cause the one or more processor(s) 121 to perform operations, such as any of the operations and functions for which the computing device(s) 120 are configured, the operations for sensor planning (e.g., method 300), as described herein, the operations for defining or receiving an area of interest, the operations for identifying or receiving one or more sensing parameters for the one or more sensors, the operations for determining a sampling combination for acquiring a plurality of samples by the one or more sensors based at least in part on the one or more sensing parameters, the operations for acquiring the plurality of samples of the area of interest using the one or more sensors based at least on the sampling combination, and/or any other operations or functions of the one or more computing device(s) 120. The instructions 123 can be software written in any suitable programming language or can be implemented in hardware. Additionally, and/or alternatively, the instructions 123 can be executed in logically and/or virtually separate threads on processor(s) 121. The memory device(s) 122 can further store data 124 that can be accessed by the processor(s) 121. For example, the data 124 can include the one or more of the samples, the sampling combinations, the sensing parameters, the defined area of interest, the score function, the RL algorithm, the combinations of overlap exponentials, and/or any other data and/or information described herein.

[0034] The computing device(s) 120 can also include a network interface 125 used to communicate, for example, with the other components of apparatus 90 (e.g., via network 410). The network interface 125 can include any suitable components for interfacing with one or more network(s), including for example, transmitters, receivers, ports, controllers, antennas, and/or other suitable components.

[0035] The technology discussed herein makes reference to computer-based systems and actions taken by and information sent to and from computer-based systems. One of ordinary skill in the art will recognize that the inherent flexibility of computer-based systems allows for a great variety of possible configurations, combinations, and divisions of tasks and functionality between and among components. For instance, processes discussed herein can be implemented using a single computing device or multiple computing devices working in combination. Databases, memory, instructions, and applications can be implemented on a single system or distributed across multiple systems. Distributed components can operate sequentially or in parallel.

[0036] This written description uses examples to disclose the invention, including the best mode, and also to enable any person skilled in the art to practice the invention, including making and using any devices or systems and performing any incorporated methods. The patentable scope of the invention is defined by the claims.


Claims

1. A computer-implemented method of sensor planning (300) for acquiring samples via an apparatus (90) including one or more sensors (110), the computer-implemented method (300) comprising:

defining (310), by one or more computing devices (120), an area of interest (130);

identifying (320), by the one or more computing devices (120), one or more sensing parameters for the one or more sensors (110);

determining (330), by the one or more computing devices (120), one or more sampling combinations for acquiring a plurality of samples of the area of interest by the one or more sensors (110) based at least in part on the one or more sensing parameters, comprising:

determining (332), by one or more computing devices (120), a combination of overlap exponentials based at least on a reinforcement learning algorithm, wherein an overlap exponential is a factor by which overlap between the sample and a previous sample is encouraged;

calculating (334), by one or more computing devices (120), a score function for the one or more sampling combinations based at least on a total area covered by the one or more sensors (110), an overlap perimeter, and the combination of overlap exponentials, wherein an overlap perimeter is a quantity of the sample that is redundant relative to a previous sample; and

selecting (336), by one or more computing devices (120), the sampling combination corresponding to a maximum score function; and

providing (340), by the one or more computing devices (120), one or more command control signals to the apparatus (90) including the one or more sensors (110) to acquire the plurality of samples of the area of interest (130) using the one or more sensors (110) based at least on the sampling combination.


 
2. The computer-implemented method (300) of claim 1, wherein the reinforcement learning algorithm comprises using at least one of a SARSA, Q-Learning, and Policy Gradient reinforcement learning algorithm.
 
3. The computer-implemented method (300) of claim 1, wherein determining a combination of overlap exponentials based at least on a reinforcement learning algorithm comprises determining combinations of zero and non-zero overlap exponentials.
 
4. The computer-implemented method (300) of claim 1, wherein determining a sampling combination for acquiring a plurality of samples by the one or more sensors (110) is based at least on a score function, wherein the score function is a function of at least one of a total area covered, an overlap perimeter, and an overlap exponential.
 
5. The computer-implemented method (300) of claim 1, wherein identifying one or more sensing parameters includes calculating a curvature and/or normal vector of at least a portion (131) of the area of interest (130).
 
6. The computer-implemented method (300) of claim 1, wherein identifying one or more sensing parameters includes defining a total area covered by the one or more sensors (110).
 
7. The computer-implemented method (300) of claim 1, wherein identifying one or more sensing parameters includes defining one or more of a measurement resolution, a field of view, and/or a depth of field.
 
8. The computer-implemented method (300) of claim 1, wherein defining an area of interest includes receiving a point cloud.
 
9. The computer-implemented method (300) of claim 1, wherein defining an area of interest (130) includes defining a finite space in which the one or more sensors (110) operates.
 
10. The computer-implemented method (300) of claim 1, wherein acquiring the plurality of samples includes translating one or more sensors (110) and/or translating the area of interest (130).
 
11. The computer-implemented method (300) of claim 1, wherein identifying one or more sensing parameters includes defining one or more translations of the one or more sensors (110).
 
12. The computer-implemented method (300) of claim 1, wherein the one or more sensors (110) is an imaging device, a proximity sensor, or a combination thereof.
 
13. A robotic sensing apparatus (90) for sensor planning, the apparatus (90) comprising one or more sensors (110) and a computing device (120) comprising one or more processors (121) and one or more memory devices (122), the one or more memory devices (122) storing instructions (123) that when executed by the one or more processors (121) cause the one or more processors (121) to perform operations, the operations comprising:

receiving an area of interest;

receiving one or more sensing parameters for the one or more sensors;

determining one or more sampling combinations for acquiring a plurality of samples of the area of interest by the one or more sensors, comprising:

determining a combination of overlap exponentials based at least on a reinforcement learning algorithm, comprising determining combinations of zero and non-zero overlap exponentials, wherein an overlap exponential is a factor by which overlap between the sample and a previous sample is encouraged;

calculating a score function for the one or more sampling combinations based at least on a total area covered by the one or more sensors, an overlap perimeter, and the combination of overlap exponentials, wherein an overlap perimeter is a quantity of the sample that is redundant relative to a previous sample; and

selecting the sampling combination corresponding to a maximum score function; and

acquiring the plurality of samples of the area of interest using the one or more sensors based at least on the sampling combination.


 
14. The apparatus of claim 13, wherein determining a combination of overlap exponentials based at least on a reinforcement learning algorithm includes using a SARSA, Q-Learning, or Policy Gradient reinforcement learning algorithm.
 
15. The apparatus of claim 13, wherein determining a combination of overlap exponentials based at least on a reinforcement learning algorithm includes determining combinations of zero and non-zero overlap exponentials.
 


Ansprüche

1. Computerimplementiertes Verfahren zur Sensorplanung (300) zum Erfassen von Abtastwerten über eine Einrichtung (90), die einen oder mehrere Sensoren (110) enthält, wobei das computerimplementierte Verfahren (300) umfasst:

Definieren (310) eines interessierenden Bereichs (130) durch eine oder mehrere Rechenvorrichtungen (120);

Identifizieren (320) eines oder mehrerer abgetasteter Parameter für den einen oder mehrere Sensoren (110) durch die eine oder mehrere Rechenvorrichtungen (120);

Bestimmen (330) durch die eine oder mehrere Rechenvorrichtungen (120) einer oder mehrerer Abtastkombinationen zum Erfassen mehrerer Abtastwerte des interessierenden Bereichs durch den einen oder mehrere Sensoren (110), die zumindest teilweise auf dem einen oder mehreren abgetasteten Parametern basieren, umfassend:

Bestimmen (332) durch eine oder mehrere Rechenvorrichtungen (120) einer Kombination von exponentiellen Überlappungen basierend auf mindestens einem Verstärkungslernalgorithmus, wobei eine exponentielle Überlappung ein Faktor ist, durch den eine Überlappung zwischen dem Abtastwert und einem vorherigen Abtastwert begünstigt wird;

Berechnen (334) durch eine oder mehrere Rechenvorrichtungen (120) einer Bewertungsfunktion für die eine oder mehrere Abtastkombinationen basierend auf mindestens einer Gesamtfläche, die von dem einen oder mehreren Sensoren (110) abgedeckt wird, einem Überlappungsumfang und der Kombination von exponentiellen Überlappungen, wobei ein Überlappungsumfang eine Menge des Abtastwerts ist, die bezogen auf einen vorherigen Abtastwert redundant ist; und

Auswählen (336) durch eine oder mehrere Rechenvorrichtungen (120) der Abtastkombination, die einer maximalen Bewertungsfunktion entspricht; und Bereitstellen (340) durch die eine oder mehrere Rechenvorrichtungen (120) eines oder mehrerer Befehlssteuersignale an die Einrichtung (90), einschließlich des einen oder mehrerer Sensoren (110), um mehrere Abtastwerte des interessierenden Bereichs (130) unter Verwenden des einen oder mehrerer Sensoren (110) zu erfassen, mindestens basierend auf der Abtastkombination.


 
2. Computerimplementiertes Verfahren (300) nach Anspruch 1, wobei der Verstärkungslernalgorithmus die Verwendung eines SARSA-, Q-Leaming- und/oder Policy-Gradient-Verstärkungslernalgorithmus umfasst.
 
3. Computerimplementiertes Verfahren (300) nach Anspruch 1, wobei das Bestimmen einer Kombination von exponentiellen Überlappungen basierend auf mindestens einem Verstärkungslernalgorithmus das Bestimmen von Kombinationen von Überlappungsexponentialen von Null und Nicht-Null umfasst.
 
4. Computerimplementiertes Verfahren (300) nach Anspruch 1, wobei das Bestimmen einer Abtastkombination zum Erfassen mehrerer Abtastwerte durch den einen oder mehrere Sensoren (110) mindestens auf einer Bewertungsfunktion basiert, wobei die Bewertungsfunktion eine Funktion der insgesamt abgedeckten Fläche, des Überlappungsumfangs und/oder der exponentiellen Überlappung ist.
 
5. Computerimplementiertes Verfahren (300) nach Anspruch 1, wobei das Identifizieren eines oder mehrerer abgetasteter Parameter das Berechnen einer Krümmung und/oder eines Normalenvektors mindestens eines Abschnitts (131) des interessierenden Bereichs (130) beinhaltet.
 
6. Computerimplementiertes Verfahren (300) nach Anspruch 1, wobei das Identifizieren eines oder mehrerer abgetasteter Parameter das Definieren einer Gesamtfläche beinhaltet, die von dem einen oder mehreren Sensoren (110) abgedeckt wird.
 
7. Computerimplementiertes Verfahren (300) nach Anspruch 1, wobei das Identifizieren eines oder mehrerer abgetasteter Parameter das Definieren einer oder mehrerer Messauflösungen, eines Sichtfelds und/oder einer Schärfentiefe beinhaltet.
 
8. Computerimplementiertes Verfahren (300) nach Anspruch 1, wobei das Definieren eines interessierenden Bereichs das Empfangen einer Punktwolke beinhaltet.
 
9. Computerimplementiertes Verfahren (300) nach Anspruch 1, wobei das Definieren eines interessierenden Bereichs (130) das Definieren eines endlichen Raums beinhaltet, in dem der eine oder mehrere Sensoren (110) arbeiten.
 
10. Computerimplementiertes Verfahren (300) nach Anspruch 1, wobei das Erfassen mehrerer Abtastwerte das Umwandeln eines oder mehrerer Sensoren (110) und/oder das Umwandeln des interessierenden Bereichs (130) beinhaltet.
 
11. Computerimplementiertes Verfahren (300) nach Anspruch 1, wobei das Identifizieren eines oder mehrerer abgetasteter Parameter das Definieren einer oder mehrerer Umwandlungen des einen oder der mehreren Sensoren (110) beinhaltet.
 
12. Computerimplementiertes Verfahren (300) nach Anspruch 1, wobei der eine oder mehrere Sensoren (110) eine Bildgebungsvorrichtung, ein Näherungssensor oder eine Kombination davon sind.
 
13. Robotersensoreinrichtung (90) zum Sensorplanen, wobei die Einrichtung (90) einen oder mehrere Sensoren (110) und eine Rechenvorrichtung (120) umfasst, die einen oder mehrere Prozessoren (121) und einen oder mehrere Speichervorrichtungen (122) umfasst, wobei die eine oder mehrere Speichervorrichtungen (122) Anweisungen (123) speichern, die bei Ausführung durch den einen oder mehrere Prozessoren (121) den einen oder mehrere Prozessoren (121) dazu veranlassen, Operationen auszuführen, wobei die Operationen Folgendes umfassen:

Empfangen eines interessierenden Bereichs;

Empfangen eines oder mehrerer abgetasteter Parameter für den einen oder mehrere Sensoren;

Bestimmen einer oder mehrerer Abtastkombinationen zum Erfassen mehrerer Abtastwerte des interessierenden Bereichs durch den einen oder mehrere Sensoren, umfassend:

Bestimmen einer Kombination von exponentiellen Überlappung basierend auf mindestens einem Verstärkungslernalgorithmus, umfassend das Bestimmen von Kombinationen von exponentiellen Überlappungen von Null und Nicht-Null, wobei ein Überlappungsexponential ein Faktor ist, durch den eine Überlappung zwischen dem Abtastwert und einem vorherigen Abtastwert begünstigt wird;

Berechnen einer Bewertungsfunktion für die eine oder mehrere Abtastkombinationen basierend auf mindestens einer Gesamtfläche, die von dem einen oder mehreren Sensoren abgedeckt wird, einem Überlappungsumfang oder Kombination von exponentiellen Überlappungen, wobei ein Überlappungsumfang eine Menge des Abtastwerts ist, die bezogen auf einen vorherigen Abtastwert redundant ist; und

Auswählen der Abtastkombination, die einer maximalen Bewertungsfunktion entspricht; und

Erfassen mehrerer Abtastwerte des interessierenden Bereichs unter Verwenden des einen oder mehrerer Sensoren mindestens basierend auf der Abtastkombination.


 
14. Einrichtung nach Anspruch 13, wobei das Bestimmen einer Kombination von exponentiellen Überlappungen mindestens basierend auf einem Verstärkungslernalgorithmus das Verwenden eines SARSA-, Q-Leaming- und Policy-Gradient-Verstärkungslernalgorithmus beinhaltet.
 
15. Einrichtung nach Anspruch 13, wobei das Bestimmen einer Kombination von exponentiellen Überlappungen mindestens basierend auf einem Verstärkungslernalgorithmus das Bestimmen von Kombinationen von Null- und Nicht-Null exponentiellen Überlappungen beinhaltet.
 


Revendications

1. Procédé de planification de capteur (300) mis en œuvre par ordinateur pour acquérir des échantillons par l'intermédiaire d'un appareil (90) comportant un ou plusieurs capteurs (110), le procédé mis en œuvre par ordinateur (300) comprenant :

la définition (310), par un ou plusieurs dispositifs informatique (120), d'une zone d'intérêt (130) ;

l'identification (320), par le ou les dispositifs informatique (120), d'un ou plusieurs paramètres de détection pour le ou les capteurs (110) ;

la détermination (330), par le ou les dispositifs informatique (120), d'une ou plusieurs combinaisons d'échantillonnage pour acquérir une pluralité d'échantillons de zone d'intérêt par le ou les capteurs (110) sur la base au moins en partie du ou des plusieurs paramètres de détection, comprenant :

la détermination (332), par un ou plusieurs dispositifs informatique (120), d'une combinaison d'exponentielles de chevauchement sur la base d'au moins un algorithme d'apprentissage par renforcement, dans lequel une exponentielle de chevauchement est un facteur par lequel le chevauchement entre l'échantillon et un échantillon précédent est encouragé ;

le calcul (334), par un ou plusieurs dispositifs informatique (120), d'une fonction de score pour une ou plusieurs combinaisons d'échantillonnage sur la base d'au moins une zone totale couverte par le ou les capteurs (110), un périmètre de chevauchement et la combinaison d'exponentielles de chevauchement, dans lequel un périmètre de chevauchement est une quantité de l'échantillon qui est redondante par rapport à un échantillon précédent ; et

la sélection (336), par un ou plusieurs dispositifs informatique (120), de la combinaison d'échantillonnage correspondant à une fonction de score maximum ; et

la fourniture (340), par le ou les dispositifs informatique (120), d'un ou plusieurs signaux de commande de commande à l'appareil (90) comportant le ou les capteurs (110) pour acquérir la pluralité d'échantillons de la zone d'intérêt (130) à l'aide du ou des capteurs (110) sur la base d'au moins la combinaison d'échantillonnage.


 
2. Procédé mis en œuvre par ordinateur (300) selon la revendication 1, dans lequel l'algorithme d'apprentissage par renforcement comprend l'utilisation d'un algorithme d'apprentissage par renforcement SARSA et/ou Q-Leaming et/ou Policy Gradient.
 
3. Procédé mis en œuvre par ordinateur (300) selon la revendication 1, dans lequel la détermination d'une combinaison d'exponentielles de chevauchement sur la base d'au moins un algorithme d'apprentissage par renforcement comprend la détermination de combinaisons d'exponentielles de chevauchement nul et non nul.
 
4. Procédé mis en œuvre par ordinateur (300) selon la revendication 1, dans lequel la détermination d'une combinaison d'échantillonnage pour acquérir une pluralité d'échantillons par le ou les capteurs (110) est basée au moins sur une fonction de score, dans lequel la fonction de score est une fonction d'une zone totale couverte et/ou un périmètre de chevauchement et/ou une exponentielle de chevauchement.
 
5. Procédé mis en œuvre par ordinateur (300) selon la revendication 1, dans lequel l'identification d'un ou plusieurs paramètres de détection comporte le calcul d'une courbure et/ou d'un vecteur normal d'au moins une partie (131) de la zone d'intérêt (130).
 
6. Procédé mis en œuvre par ordinateur (300) selon la revendication 1, dans lequel l'identification d'un ou plusieurs paramètres de détection comporte la définition d'une zone totale couverte par le ou les capteurs (110).
 
7. Procédé mis en œuvre par ordinateur (300) selon la revendication 1, dans lequel l'identification d'un ou plusieurs paramètres de détection comporte la définition d'une résolution de mesure et/ou d'un champ de vision, et/ou d'une profondeur de champ.
 
8. Procédé (300) mis en œuvre par ordinateur selon la revendication 1, dans lequel la définition d'une zone d'intérêt comporte la réception d'un nuage de points.
 
9. Procédé mis en œuvre par ordinateur (300) selon la revendication 1, dans lequel la définition d'une zone d'intérêt (130) comporte la définition d'un espace fini dans lequel le ou les capteurs (110) fonctionnent.
 
10. Procédé mis en œuvre par ordinateur (300) selon la revendication 1, dans lequel l'acquisition de la pluralité d'échantillons comporte la translation d'un ou plusieurs capteurs (110) et/ou la translation de la zone d'intérêt (130).
 
11. Procédé mis en œuvre par ordinateur (300) selon la revendication 1, dans lequel l'identification d'un ou plusieurs paramètres de détection comporte la définition d'une ou plusieurs translations du ou des capteurs (110).
 
12. Procédé (300) mis en œuvre par ordinateur selon la revendication 1, dans lequel le ou les capteurs (110) sont un dispositif d'imagerie, un capteur de proximité ou une combinaison de ceux-ci.
 
13. Appareil de détection robotique (90) pour la planification de détection, l'appareil (90) comprenant un ou plusieurs capteurs (110) et un dispositif informatique (120) comprenant un ou plusieurs processeurs (121) et un ou plusieurs dispositifs de mémoire (122), le ou les dispositifs de mémoire (122) stockant des instructions (123) qui, lorsqu'elles sont exécutées par un ou plusieurs processeurs (121), amènent le processeur à effectuer des opérations (121), les opérations comprenant :

la réception d'une zone d'intérêt ;

la réception d'un ou plusieurs paramètres de détection pour le ou les capteurs ;

la détermination d'une ou plusieurs combinaisons d'échantillonnage pour acquérir une pluralité d'échantillons de la zone d'intérêt par le ou les capteurs, comprenant :

la détermination d'une combinaison d'exponentielles de chevauchement sur la base d'au moins un algorithme d'apprentissage par renforcement, comprenant la détermination de combinaisons d'exponentielles de chevauchement nul et non nul, dans lequel une exponentielle de chevauchement est un facteur par lequel le chevauchement entre l'échantillon et un échantillon précédent est encouragé ;

le calcul d'une fonction de score pour la ou les combinaisons d'échantillonnage sur la base d'au moins une zone totale couverte par le ou les capteurs, un périmètre de chevauchement et la combinaison d'exponentielles de chevauchement, dans lequel un périmètre de chevauchement est une quantité de l'échantillon qui est redondante par rapport à un échantillon précédent ; et

la sélection de la combinaison d'échantillonnage correspondant à une fonction de score maximum ; et

l'acquisition de la pluralité d'échantillons de la zone d'intérêt à l'aide du ou des capteurs sur la base d'au moins la combinaison d'échantillonnage.


 
14. Appareil selon la revendication 13, dans lequel la détermination d'une combinaison d'exponentielles de chevauchement basée au moins sur un algorithme d'apprentissage par renforcement comporte l'utilisation d'un algorithme d'apprentissage par renforcement SARSA, Q-Learning ou Policy Gradient.
 
15. Appareil selon la revendication 13, dans lequel la détermination d'une combinaison d'exponentielles de chevauchement sur la base au moins d'un algorithme d'apprentissage par renforcement comporte la détermination de combinaisons d'exponentielles de chevauchement nul et non nul.
 




Drawing














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.

Non-patent literature cited in the description