Rapid robotic imitation learning of force-torque tasks

Patent metadata

Field	Value
Publication number	US-9403273-B2
Application number	US-201414285867-A
Country	US
Kind code	B2
Filing date	May 23, 2014
Priority date	May 23, 2014
Publication date	Aug 2, 2016
Grant date	Aug 2, 2016

Abstract

A method of training a robot to autonomously execute a robotic task includes moving an end effector through multiple states of a predetermined robotic task to demonstrate the task to the robot in a set of n training demonstrations. The method includes measuring training data, including at least the linear force and the torque via a force-torque sensor while moving the end effector through the multiple states. Key features are extracted from the training data, which is segmented into a time sequence of control primitives. Transitions between adjacent segments of the time sequence are identified. During autonomous execution of the same task, a controller detects the transitions and automatically switches between control modes. A robotic system includes a robot, force-torque sensor, and a controller programmed to execute the method.

First claim

The invention claimed is:

1. A method for training a robot to autonomously execute a predetermined robotic task requiring application of a linear force and a torque to an object by an end effector of the robot, the method comprising:
moving the end effector through multiple states of the predetermined robotic task to thereby demonstrate the predetermined robotic task to the robot in a set of n training demonstrations;
measuring a set of training data, including measuring at least the linear force and the torque via a force-torque sensor, while moving the end effector through the multiple states of the predetermined robotic task;
extracting key features from the measured set of training data via a controller, including segmenting the measured set of training data into a time sequence of control primitives and identifying transitions between adjacent segments of the time sequence;
measuring the linear force and the torque via the force-torque sensor as online/real-time data during a subsequent autonomous execution of the demonstrated robotic task by the robot;
detecting transitions via the controller during the subsequent autonomous execution of the demonstrated robotic task by the robot; and
automatically switching between a plurality of different control modes during the subsequent autonomous execution of the demonstrated robotic task in response to the detected transitions.

2. The method of claim 1 , wherein moving an end effector includes backdriving the end effector.

3. The method of claim 1 , wherein moving an end effector includes commanding the end effector via a user input device.

4. The method of claim 1 , wherein n≦5.

5. The method of claim 4 , wherein n≦2.

6. The method of claim 1 , wherein detecting the transitions includes detecting a transition between each of a position control primitive, a hybrid force control primitive, and a goal force control primitive, and wherein segmenting the measured set of training data into a time sequence of control primitives includes segmenting the measured set of training data into combinations of the position control primitive, the hybrid force control primitive, and the goal force control primitive.

7. The method of claim 1 , wherein detecting the transitions includes executing a scoring function via the controller to calculate a probability that a data point in the online/real-time data during the subsequent autonomous execution of the demonstrated robotic task represents one of the transitions.

8. A robotic system comprising:
a robot having an end effector;
at least one force-torque sensor positioned with respect to the robot and operable to measure linear and rotational forces applied to an object by the end effector; and
a controller in communication with the robot and the at least one force-torque sensor, and having a processor and memory on which is recorded instructions for training the robot to autonomously execute a robotic task requiring an application of linear forces and torques to the object by the end effector, wherein the controller is configured to execute the instructions to cause the controller to:
record a movement of the end effector through multiple states of the robotic task during a set of n demonstrations of the robotic task;
measure a set of training data, including measuring, via the at least one force-torque sensor, a linear force and a torque applied to the object by the robot while the end effector moves through the multiple states;
extract key features from the measured set of training data, including segmenting the measured set of training data into a time sequence of control primitives separated and identifying transitions between the segments of the time sequence;
measure at least the linear force and the torque via the force-torque sensor as online/real-time data during a subsequent autonomous execution of the demonstrated robotic task by the robot;
detect the transitions during the subsequent autonomous execution of the demonstrated robotic task by the robot; and
switch between a plurality of different control modes in response to the detected transitions during the subsequent autonomous execution of the demonstrated robotic task.

9. The robotic system of claim 8 , wherein the controller is configured to record the movement of the end effector by recording motion imparted by a backdriving of the end effector.

10. The robotic system of claim 8 , further comprising a user input device programmed to command the end effector to move during the set of n task demonstrations.

11. The robotic system of claim 8 , wherein the robot includes a wrist, and wherein the force-torque sensor is embedded in the wrist.

12. The robotic system of claim 8 , wherein the control primitives include a position control primitive, a hybrid force control primitive, and goal force control primitive, and wherein the controller is programmed to detect a corresponding transition between each of the control primitives.

13. The robotic system of claim 8 , wherein the controller detects the transitions by executing a scoring function to thereby calculate a probability that a data point in the online/real-time data during the subsequent autonomous execution of the demonstrated task represents one of the transitions.

14. A method for training a robot to autonomously execute a robotic task requiring application of a linear force and a torque to an object by an end effector of the robot, the method comprising:
backdriving the end effector through multiple states of a predetermined robotic task to thereby demonstrate the predetermined robotic task to the robot in a set of n training demonstrations, wherein n≦3;
measuring a set of training data, including measuring at least the linear force and the torque via a force-torque sensor embedded in a wrist of the robot, while backdriving the end effector through the multiple states of the predetermined robotic task;
extracting key features from the measured set of training data via a controller, including segmenting the measured set of training data into a time sequence of control primitives using a rolling average of the measured set of training data and identifying transitions between adjacent segments of the time sequence;
measuring at least the linear force and the torque via the force-torque sensor as online/real-time data during a subsequent autonomous execution of the demonstrated robotic task by the robot;
detecting the transitions in the online/real-time data during the subsequent autonomous execution of the demonstrated robotic task by the robot, including detecting a transition between each of a position control primitive, a hybrid force control primitive, and a goal force control primitive; and
automatically switching between a plurality of different control modes in response to the detected transitions.

15. The method of claim 14 , wherein detecting the transitions includes executing a scoring function that calculates a probability that a data point in the online/real-time data represents one of the transitions.

Assignees

• Gm Global Tech Operations Llc

Inventors

• Payton David W
• Uhlenbrock Ryan M
• Ku Li Yang

Classifications

• G05B19/423

  Teaching successive positions by walk-through, i.e. the tool head or end effector being grasped and guided directly, with or without servo-assistance, to follow a path · CPC title

• G05B2219/36442

  Automatically teaching, teach by showing · CPC title

• B25J9/1664 Primary

  characterised by motion, path, trajectory planning · CPC title

• Y10S901/03

  Teaching system · CPC title

• B25J9/163 Primary

  learning, adaptive, model based, rule based expert control · CPC title