Atkeson, C. G. and S. Schaal (1997). Robot Learning From Demonstration, Machine Learning: Proceedings of the Fourteenth International Conference (ICML '97), Edited by Douglas H. Fisher, Jr. pp. 12-20, Morgan Kaufmann, San Francisco, CA, 1997.

@InProceedings{,
  author = 	 {Christopher. G. Atkeson, Stefan Schaal},
  title = 	 {Robot learning from demonstration},
  booktitle = 	 {Machine Learning:
Proceedings of the Fourteenth International Conference (ICML '97)},
  OPTpages = 	 {12--20},
  OPTyear = 	 {1997},
  OPTeditor = 	 {Douglas H. Fisher, Jr.},
  OPTpublisher = {Morgan Kaufmann, San Francisco, CA},
}

Author of the summary: Jim R. Davies, 2000, jim@jimdavies.org

The actual paper is online at http://www.cc.gatech.edu/fac/Chris.Atkeson/publications.html.

Cite this paper for:

• mimicking actions is not sufficient for this task
• a model-based planning process supports rapid learning
• Building complex motions out of motion primitives
• learning a policy through demonstration

A computer vision system watched a human do the task. The things to watch for were put in by the experimenters: pendulum angle, angular velocity, hand position, hand velocity.

Building complex motions out of motion primitives.

The system tries to imitate the human, and rewards or punishes itself, as appropriate, depending on how much what it did looked like that the human did.

The motions of the robot hand were very different than that demonstrated. This is interesting: from demonstration it learned a new way to do something.

Summary author's notes:

• gritty details excluded from this summary.