Real time robot collision avoidance

• US 10,131,053 B1
• Filed: 09/14/2016
• Issued: 11/20/2018
• Est. Priority Date: 09/14/2016
• Status: Active Grant

• Current Assignee: Google LLC (Alphabet Inc.)
• Original Assignee: X Development LLC (Alphabet Inc.)
• Inventors: Sampedro, Peter Pastor, Nagarajan, Umashankar
• Primary Examiner(s): Sample, Jonathan L
• Application Number:
  US15/265,547
• Time in Patent Office: 797 Days
• Field of Search: 700245-264
• CPC Class Codes: B25J 9/1666 Avoiding collision or forbidden zones
• CPC Class Codes: G05B 2219/40476 Collision, planning for collision free path

Abstract

Methods and apparatus related to robot collision avoidance. One method may include: receiving robot instructions to be performed by a robot; at each of a plurality of control cycles of processor(s) of the robot: receiving trajectories to be implemented by actuators of the robot, wherein the trajectories define motion states for the actuators of the robot during the control cycle or a next control cycle, and wherein the trajectories are generated based on the robot instructions; determining, based on a current motion state of the actuators and the trajectories to be implemented, whether implementation of the trajectories by the actuators prevents any collision avoidance trajectory from being achieved; and selectively providing the trajectories or collision avoidance trajectories for operating the actuators of the robot during the control cycle or the next control cycle depending on a result of the determining.

Claims

1. A method for learning a robot'"'"'s capabilities comprising:

receiving robot instructions to be performed by the robot, wherein the robot instructions are randomly generated to cause the robot to perform random movements;

at each of a plurality of control cycles of one or more processors of the robot;

receiving trajectories to be implemented by actuators of the robot, wherein the trajectories define motion states for the actuators of the robot during the control cycle or a next control cycle, and wherein the trajectories are generated based on the robot instructions;

determining, based on a current motion state of the actuators and the trajectories to be implemented, whether implementation of the trajectories by the actuators prevents any collision avoidance trajectory of a plurality of collision avoidance trajectories calculated for the actuators during an immediately preceding control cycle from being achieved;

when implementation of the trajectories by the actuators does not prevent any collision avoidance trajectory from being achieved, providing the trajectories for operating the actuators of the robot during the control cycle or the next control cycle;

when implementation of the trajectories prevents any collision avoidance trajectory from being achieved, providing the plurality of collision avoidance trajectories for operating the actuators of the robot during the control cycle or the next control cycle, without implementing the trajectories;

operating the actuators of the robot during the control cycle or the next control cycle based on the trajectories or the collision avoidance trajectories to generate data; and

training a machine learning model used to control the robot based on the data generated during the operating.

2. The method of claim 1, wherein determining whether implementation of the trajectories prevents any collision avoidance trajectories from being achieved is based on projecting one or more future motion states of the actuators based on the current motion state.

3. The method of claim 1, wherein receiving the trajectories includes receiving, during the control cycle, trajectories to be implemented by the actuators during the next control cycle.

4. The method of claim 1, wherein the trajectories are generated based on output of a neural network model.

5. The method of claim 1, wherein the collision avoidance trajectories comprise safe stopping trajectories achievable by the actuators of the robot.

6. The method of claim 1, further comprising receiving environmental parameters that indicate one or more properties of environmental objects that are external to the robot, wherein determining whether implementation of the trajectories by the actuators prevents any collision avoidance trajectories from being achieved is based on the environmental parameters.

7. The method of claim 1, wherein the robot is a one of a plurality of robots, the method further comprising:

operating each of the plurality of robots based on the robot instructions;

storing, in one or more computer readable media, data generated by the plurality of robots during the operation of the plurality of robots based on the robot instructions; and

providing at least some of the data or additional data generated based on the data.

8. The method of claim 7, wherein the data generated by the plurality of robots includes data indicative of one or more determinations that collision avoidance trajectories were provided for operating actuators of one or more of the plurality of robots.

9. The method of claim 8, wherein the additional data generated based on the data comprises updated robot instructions.

10. The method of claim 7, further comprising:

configuring a simulated operating space of a robot simulator based on received environmental parameters;

operating the robot simulator based on the robot instructions with the simulated operating space configured based on the received environmental parameters; and

storing, in one or more of the computer readable media, simulated data generated by the robot simulator during the operation of the robot simulator based on the robot instructions;

wherein providing at least some of the data further comprises providing at least some of the simulated data or the additional data further based on the simulated data.

11. A system comprising one or more processors and memory operably coupled with the one or more processors, wherein the memory stores instructions that, in response to execution of the one or more processors, cause the one or more processors to:

receive robot instructions to be performed by a robot, wherein the robot instructions are randomly generated to cause the robot to perform random movements;

at each of a plurality of control cycles of one or more processors of the robot;

receive trajectories to be implemented by actuators of the robot, wherein the trajectories define motion states for the actuators of the robot during the control cycle or a next control cycle, and wherein the trajectories are generated based on the robot instructions;

determine, based on a current motion state of the actuators and the trajectories to be implemented, whether implementation of the trajectories by the actuators prevents any collision avoidance trajectory of a plurality of collision avoidance trajectories calculated for the actuators during an immediately preceding control cycle from being achieved;

when implementation of the trajectories by the actuators does not prevent any collision avoidance trajectory from being achieved, provide the trajectories for operating the actuators of the robot during the control cycle or the next control cycle;

when implementation of the trajectories prevents any collision avoidance trajectory from being achieved, provide collision avoidance trajectories for operating the actuators of the robot during the control cycle or the next control cycle, without implementing the trajectories;

operate the actuators of the robot during the control cycle or the next control cycle based on the trajectories or the collision avoidance trajectories to generate data; and

train a machine learning model used to control the robot based on the data generated during the operating.

12. The system of claim 11, wherein the determination of whether implementation of the trajectories prevents any collision avoidance trajectories from being achieved is based on a projection of one or more future motion states of the actuators during one or more future control cycles.

13. The system of claim 11, wherein the instructions include instructions to receive trajectories to be implemented by the actuators during the next control cycle during the current control cycle.

14. A method comprising:

receiving, by one or more processors, robot instructions to be performed by a robot, wherein the robot instructions are randomly generated to cause the robot to perform random movements;

determining, by one or more of the processors, for each iteration of a sequence of motion states traversed by one or more actuators of the robot based on the robot instructions, collision avoidance trajectories achievable by actuators of the robot from a future motion state;

while the one or more actuators of the robot are in a given motion state of the sequence of motion states, determining, by one or more of the processors, that collision avoidance trajectories of a plurality of collision avoidance trajectories calculated for the actuators during an immediately preceding control cycle are not achievable by the actuators in a future iteration of the sequence of motion states;

operating, by one or more of the processors, the robot to implement a collision avoidance trajectory associated with the given motion state in response to determining that the collision avoidance trajectories are not achievable in the future iteration, wherein the operating causes data to be generated; and

training a machine learning model used to control the robot based on the data generated during the operating.

15. The method of claim 14, wherein the collision avoidance trajectories comprise safe stopping trajectories achievable by the actuators of the robot.

16. The method of claim 14, further comprising projecting, by one or more of the processors, for each iteration of the sequence of motion states traversed by the one or more actuators of the robot based on the robot instructions, based on a current motion state and one or more instructions of the robot instructions, the future motion state of the one or more actuators.