Institute Homepage DE Sign In

Back

Autonomous Motion Conference Paper 2001

Learning inverse kinematics

Thumb ticker sm ss
Autonomous Motion
Stefan Schaal
  • Director

Real-time control of the endeffector of a humanoid robot in external coordinates requires computationally efficient solutions of the inverse kinematics problem. In this context, this paper investigates learning of inverse kinematics for resolved motion rate control (RMRC) employing an optimization criterion to resolve kinematic redundancies. Our learning approach is based on the key observations that learning an inverse of a non uniquely invertible function can be accomplished by augmenting the input representation to the inverse model and by using a spatially localized learning approach. We apply this strategy to inverse kinematics learning and demonstrate how a recently developed statistical learning algorithm, Locally Weighted Projection Regression, allows efficient learning of inverse kinematic mappings in an incremental fashion even when input spaces become rather high dimensional. The resulting performance of the inverse kinematics is comparable to Liegeois ([1]) analytical pseudo inverse with optimization. Our results are illustrated with a 30 degree-of-freedom humanoid robot.

Author(s): D’Souza, A. and Vijayakumar, S. and Schaal, S.
Book Title: IEEE International Conference on Intelligent Robots and Systems (IROS 2001)
Year: 2001
Publisher: Piscataway, NJ: IEEE
Bibtex Type: Conference Paper (inproceedings)
Address: Maui, Hawaii, Oct.29-Nov.3
URL: http://www-clmc.usc.edu/publications/D/dsouza-IROS2001.pdf
Cross Ref: p1454
Electronic Archiving: grant_archive
Note: clmc

BibTex 

@inproceedings{DSouza_IICIRS_2001,
  title = {Learning inverse kinematics},
  booktitle = {IEEE International Conference on Intelligent Robots and Systems (IROS 2001)},
  abstract = {Real-time control of the endeffector of a humanoid robot in external coordinates requires computationally efficient solutions of the inverse kinematics problem. In this context, this paper investigates learning of inverse kinematics for resolved motion rate control (RMRC) employing an optimization criterion to resolve kinematic redundancies. Our learning approach is based on the key observations that learning an inverse of a non uniquely invertible function can be accomplished by augmenting the input representation to the inverse model and by using a spatially localized learning approach. We apply this strategy to inverse kinematics learning and demonstrate how a recently developed statistical learning algorithm, Locally Weighted Projection Regression, allows efficient learning of inverse kinematic mappings in an incremental fashion even when input spaces become rather high dimensional. The resulting performance of the inverse kinematics is comparable to Liegeois ([1]) analytical pseudo inverse with optimization. Our results are illustrated with a 30 degree-of-freedom humanoid robot.},
  publisher = {Piscataway, NJ: IEEE},
  address = {Maui, Hawaii, Oct.29-Nov.3},
  year = {2001},
  note = {clmc},
  slug = {dsouza_iicirs_2001},
  author = {D'Souza, A. and Vijayakumar, S. and Schaal, S.},
  crossref = {p1454},
  url = {http://www-clmc.usc.edu/publications/D/dsouza-IROS2001.pdf}
}