Evolutionary imitation learning

1. A method, comprising:

training an artificial neural network (ANN), using a supervised machine learning technique, according to first example data representative of a behavior to be imitated by the ANN in performing a task;

generating, using the ANN, first sample data representative of a behavior of the ANN in performing the task;

generating second sample data from the first sample data according to a criterion configured to select mutations of the behavior of the ANN; and

training the ANN according to the second sample data using the supervised machine learning technique.

2. The method of claim 1 , wherein the generating of the second sample data comprises:

applying first changes to aspects of the first sample data to generate third sample data; and

selecting from the aspects for applying second changes to generate the second sample data based on performance levels of the third sample data evaluated according to the criterion.

3. The method of claim 2 , wherein the generating of the second sample data further comprises:

selecting the second sample data based on comparing a performance level of the first example data and a performance level of sampled data mutated from the first sample data.

4. The method of claim 3 , further comprising:

evaluating a performance level of a robot in each instance of performing in the first example data, in the first sample data, and in the second sample data.

5. The method of claim 4 , wherein the ANN is configured to receive input data representative of sensor data received during the robot performing a respective task and to generate output data representative of control signals applied to the robot during performing the respective task.

6. The method of claim 5 , wherein the first example data includes first input data to the ANN and first output data used to control the robot; the training according to the first example data reduces differences between the first output data and outputs generated by the ANN responsive to the first input data.

7. The method of claim 6 , wherein the second sample data includes second input data to the ANN and second output data; and the training according to the second sample data uses the second sample data as second example data to reduce differences between the second output data and outputs generated by the ANN responsive to the second input data.

8. The method of claim 7 , wherein the first example data includes data representative of an example of control signals selected by a human operator to operate the robot to perform the task.

9. The method of claim 8 , wherein the second sample data is generated from the first sample data without a human operator controlling the robot.

10. The method of claim 9 , wherein the generating of the second sample data is performed using a technique of evolutionary algorithm.

11. The method of claim 10 , wherein the first example data is collected during performing the task in a first environment; and the second sample data is generated in performing the task by the robot in a second environment different from the first environment.

12. The method of claim 10 , wherein the first example data is collected during performing the task at a first performance level; and the second sample data is generated in performing the task at a second performance level higher than the first performance level.

13. The method of claim 10 , wherein the task is a first task; and the second sample data is generated in performing a second task having a modification from the first task.

14. A system, comprising:

memory storing instructions; and

at least one processor configured via the instructions to:

train an artificial neural network (ANN), using a supervised machine learning technique, according to first example data representative of a behavior to be imitated by the ANN in performing a task;

generate, using the ANN, first sample data representative of a behavior of the ANN in performing the task;

modify the first sample data to generate second sample data according to a criterion configured to select mutations of the behavior of the ANN; and

train the ANN according to the second sample data using the supervised machine learning technique.

15. The system of claim 14 , further comprising:

a robot having actuators and sensors to generate input data for the ANN, the ANN trained using the supervised machine learning technique to generate output data to control the actuators;

wherein the at least one processor is further configured to generate the second sample data from the first sample data using a technique of evolutionary algorithm.

16. The system of claim 15 , wherein the first example data includes first input data to the ANN and first output data used to control the actuator in performing the task; the ANN is trained according to the first example data to reduce differences between the first output data and outputs generated by the ANN responsive to the first input data; the second sample data includes second input data to the ANN and second output data; and the ANN is trained according to the second sample data to reduce differences between the second output data and outputs generated by the ANN responsive to the second input data.

17. A non-transitory computer readable medium storing instructions which, when executed by a computing system, cause the computing system to perform a method, the method comprising:

training an artificial neural network (ANN), using a supervised machine learning technique, according to first example data representative of a behavior to be imitated by the ANN in performing a task;

generating, using the ANN, first sample data representative of a behavior of the ANN in performing the task;

modifying first sample data using a technique of evolutionary algorithm to generate second sample data according to a criterion configured to select mutations of the behavior of the ANN; and

training the ANN according to the second sample data using the supervised machine learning technique.

18. The non-transitory computer readable medium of claim 17 , wherein the first example data is collected during performing the task in a first environment; and the second sample data is generated in performing the task in a second environment different from the first environment.

19. The non-transitory computer readable medium of claim 17 , wherein the first example data is collected during performing the task at a first performance level; and the second sample data is generated in performing the task at a second performance level higher than the first performance level.

20. The non-transitory computer readable medium of claim 17 , wherein the task is a first task; and the second sample data is generated in performing a second task having a modification from the first task.