Neural networks to generate robotic task demonstrations

Patent metadata

Field	Value
Publication number	US-12415270-B2
Application number	US-202217695753-A
Country	US
Kind code	B2
Filing date	Mar 15, 2022
Priority date	Dec 17, 2021
Publication date	Sep 16, 2025
Grant date	Sep 16, 2025

Abstract

A technique for training a neural network, including generating a plurality of input vectors based on a first plurality of task demonstrations associated with a first robot performing a first task in a simulated environment, wherein each input vector included in the plurality of input vectors specifies a sequence of poses of an end-effector of the first robot, and training the neural network to generate a plurality of output vectors based on the plurality of input vectors. Another technique for generating a task demonstration, including generating a simulated environment that includes a robot and at least one object, causing the robot to at least partially perform a task associated with the at least one object within the simulated environment based on a first output vector generated by a trained neural network, and recording demonstration data of the robot at least partially performing the task within the simulated environment.

First claim

What is claimed is:

1. A computer-implemented method for training a neural network to enable a robot task, the computer-implemented method comprising:
generating a plurality of input vectors based on a first plurality of task demonstrations associated with a first robot performing a first task in a simulated environment, wherein each input vector included in the plurality of input vectors specifies a sequence of past poses of an end-effector of the first robot in the simulated environment; and
training the neural network to generate a plurality of output vectors based on the plurality of input vectors to generate a trained neural network, wherein the trained neural network subsequently generates training data for training a second neural network for robot learning.

2. The computer-implemented method of claim 1 , wherein training the neural network comprises:
inputting a first input vector included in the plurality of input vectors to the neural network, wherein the first input vector specifies a first sequence of poses of the end-effector; and
training the neural network to predict a first output vector based on the first input vector, wherein the first output vector comprises a second sequence of poses of the end-effector that is subsequent to the first sequence of poses of the end-effector.

3. The computer-implemented method of claim 1 , wherein a first pose of the end-effector of the first robot comprises a position and orientation of the end-effector within the simulated environment.

4. The computer-implemented method of claim 1 , wherein each input vector included in the plurality of input vectors further specifies a sequence of grip statuses of the end-effector of the first robot.

5. The computer-implemented method of claim 1 , wherein training the neural network comprises:
inputting a first input vector included in the plurality of input vectors to the neural network, wherein the first input vector specifies a first sequence of grip statuses of the end-effector; and
training the neural network to predict a first output vector based on the first input vector, wherein the first output vector comprises a second sequence of grip statuses of the end-effector that is subsequent to the first sequence of grip statuses of the end-effector.

6. The computer-implemented method of claim 1 , wherein each input vector included in the plurality of input vectors further specifies a sequence of poses associated with at least one object in the simulated environment.

7. The computer-implemented method of claim 1 , wherein training the neural network comprises:
inputting a first input vector included in the plurality of input vectors to the neural network, wherein the first input vector specifies a first sequence of poses of at least one object in the simulated environment; and
training the neural network to predict a first output vector based on the first input vector, wherein the first output vector comprises a second sequence of poses of the at least one object in the simulated environment that is subsequent to the first sequence of poses of the at least one object in the simulated environment.

8. The computer-implemented method of claim 1 , wherein none of the input vectors included in the plurality of input vectors specifies a configuration parameter that defines a model type for the first robot.

9. The computer-implemented method of claim 1 , further comprising re-training the neural network based on a second plurality of task demonstrations associated with a second robot performing a second task in a simulated environment to generate a specialized neural network, wherein the second task is different than the first task.

10. The computer-implemented method of claim 9 , wherein the first robot is defined by a first set of configuration parameters, and the second robot is defined by a second set of configuration parameters that is different than the first set of configuration parameters.

11. One or more non-transitory computer readable media storing instructions that, when executed by one or more processors, cause the one or more processors to train a neural network to enable a robot task by performing the steps of:
generating a plurality of input vectors based on a first plurality of task demonstrations associated with a first robot performing a first task in a simulated environment, wherein each input vector included in the plurality of input vectors specifies a sequence of past poses of an end-effector of the first robot in the simulated environment; and
training the neural network to generate a plurality of output vectors based on the plurality of input vectors to generate a trained neural network, wherein the trained neural network subsequently generates training data for training a second neural network for robot learning.

12. The one or more non-transitory computer readable media of claim 11 , wherein training the neural network comprises:
inputting a first input vector included in the plurality of input vectors to the neural network, wherein the first input vector specifies a first sequence of poses of the end-effector; and
training the neural network to predict a first output vector based on the first input vector, wherein the first output vector comprises a second sequence of poses of the end-effector that is subsequent to the first sequence of poses of the end-effector.

13. The one or more non-transitory computer readable media of claim 11 , wherein each input vector included in the plurality of input vectors comprises a state vector specifying a sequence of state poses of the end-effector and an action vector specifying a sequence of action poses of the end-effector.

14. The one or more non-transitory computer readable media of claim 11 , wherein each input vector included in the plurality of input vectors further specifies a sequence of grip statuses of the end-effector of the first robot.

15. The one or more non-transitory computer readable media of claim 11 , wherein training the neural network comprises:
inputting a first input vector included in the plurality of input vectors to the neural network, wherein the first input vector specifies a first sequence of grip statuses of the end-effector; and
training the neural network to predict a first output vector based on the first input vector, wherein the first output vector comprises a second sequence of grip statuses of the end-effector that is subsequent to the first sequence of grip statuses of the end-effector.

16. The one or more non-transitory computer readable media of claim 11 , wherein each input vector included in the plurality of input vectors further specifies a sequence of poses associated with at least one object in the simulated environment.

17. The one or more non-transitory computer readable media of claim 11 , wherein training the neural network comprises:
inputting a first input vector included in the plurality of input vectors to the neural network, wherein the first input vector specifies a first sequence of poses of at least one object in the simulated environment; and
training the neural network to predict a first output vector based on the first input vector, wherein the first output vector comprises a second sequence of poses of the at least one object in the simulated environment that is subsequent to the first sequence of poses of the at least one object in the simulated environment.

18. The one or more non-transitory computer readable media of claim 11 , further comprising:
executing the neural network to generate a second plurality of task demonstrations of a seco

Assignees

• Nvidia Corp

Inventors

• Handa Ankur
• Akinola Iretiayo
• Fox Dieter
• Narang Yashraj Shyam

Classifications

• B25J9/161

  Hardware, e.g. neural networks, fuzzy logic, interfaces, processor · CPC title

• B25J9/1671

  characterised by simulation, either to verify existing program or to create and verify new program, CAD/CAM oriented, graphic oriented programming systems · CPC title

• B25J9/1612

  characterised by the hand, wrist, grip control · CPC title

• G05B2219/39298

  Trajectory learning · CPC title

• G05B2219/39271

  Ann artificial neural network, ffw-nn, feedforward neural network · CPC title