Robot with automatic selection of task-specific representations for imitation learning

• US 8,571,714 B2
• Filed: 02/22/2010
• Issued: 10/29/2013
• Est. Priority Date: 02/27/2009
• Status: Active Grant

• Current Assignee: Honda Research Institute Europe GmbH (Honda Motor Company)
• Original Assignee: Honda Research Institute Europe GmbH (Honda Motor Company)
• Inventors: Steil, Jochen, Gienger, Michael, Muehlig, Manuel
• Primary Examiner(s): MANCHO, RONNIE M
• Application Number:
  US12/709,814
• Publication Number:
  US 20100222924A1
• Time in Patent Office: 1,345 Days
• Field of Search: 700/245, 700/259, 700/254, 700/258, 700/264, 700/250, 700/257, 700/261, 700/160, 707/100, 318/568.22, 318/572, 318/571, 318/568.12, 318/568.16, 451/38, 451/177
• US Class Current: 700/258
• CPC Class Codes: G05B 19/42 Recording and playback systems, i.e. in which the programme is recorded from a cycle of operations, e.g. the cycle of operations being manually controlled, after which this record is played back on the same machine
• CPC Class Codes: G05B 2219/36442 Automatically teaching, teach by showing

Abstract

The invention proposes a method for imitation-learning of movements of a robot, wherein the robot performs the following steps:

• • observing a movement of an entity in the robot'"'"'s environment,
  • recording the observed movement using a sensorial data stream and representing the recorded movement in a different task space representations,
  • selecting a subset of the task space representations for the imitation learning and reproduction of the movement to be imitated.

Claims

1. A method for imitation-learning of movements of a robot, the method comprising:

observing a movement of an entity in the robot'"'"'s environment,

recording the observed movement using a sensorial data stream and representing the recorded movement in a plurality of task space representations,

automatically selecting, based on cues extracted from the sensorial data stream, a subset of the plurality of task space representations for the imitation learning and reproduction of the movement to be imitated, the subset comprising at least one task space representation or a sequence of different task space representations for the reproduction of the movement wherein the step of selecting a subset of the task space representations uses cues which the robot extracts from the sensorial data stream.

2. The method according to claim 1, wherein the step of selecting a subset of the task space representations uses a variance over multiple demonstrations of the movement, wherein task space representations in which the observations have a lowest variance are chosen.

3. The method according to claim 1, wherein the step of selecting a subset of the task space representations uses explicit attention generation by a human teacher.

4. The method according to claim 1, wherein the step of selecting a subset of the task space representations uses a kinematic or dynamic simulation of a human teacher.

5. The method according to claim 4, wherein task elements for the selection step are defined through discomfort of the human teacher and effort, of the human teacher during the task demonstration,

wherein the discomfort of the human teacher includes deviation from a default posture of the human teacher, and the effort of the human teacher is based on a torque of effector joint.

6. The method according to claim 1, wherein the task space selection influences the movement reproduction process on the robot.

7. The method according to claim 1, wherein an optimization method is applied on the different task space representations in order to efficiently reproduce the movement.

8. The method according to claim 1, wherein different task space representations are used sequentially over time when reproducing a learned movement.

9. A non-transitory computer readable medium for storing instructions, which, when run on a computing device of a robot, execute the method according to claim 1.

10. A robot, preferably a humanoid robot, having a computing unit designed to perform a method according to claim 1.

11. The robot of claim 10, wherein the robot is an industrial robot, which learns certain sequences of working steps by imitation learning.