Header logo is am

Learning Force Control Policies for Compliant Robotic Manipulation


Conference Paper



In this abstract, we present an approach to learning manipulation tasks on compliant robots through re- inforcement learning. We demonstrate our approach on two different manipulation tasks: opening a door with a lever door handle, and picking up a pen off the table (Fig. 1). We show that our approach can learn the force control policies required to achieve both tasks successfully. The contributions of this work are two-fold: (1) we demonstrate that learning force con- trol policies enables compliant execution of manipu- lation tasks with increased robustness as opposed to stiff position control, and (2) we introduce a policy parameterization that uses finely discretized trajectories coupled with a cost function that ensures smoothness during exploration and learning.

Author(s): Kalakrishnan, M. and Righetti, L. and Pastor, P. and Schaal, S.
Book Title: International Conference on Machine Learning (ICML)
Year: 2012

Department(s): Autonomous Motion, Movement Generation and Control
Bibtex Type: Conference Paper (inproceedings)

Cross Ref: p10519
Note: clmc
URL: http://clmc.usc.edu/publications/K/kalakrishnan-ICML2012.pdf


  title = {Learning Force Control Policies for Compliant Robotic Manipulation},
  author = {Kalakrishnan, M. and Righetti, L. and Pastor, P. and Schaal, S.},
  booktitle = {International Conference on Machine Learning (ICML)},
  year = {2012},
  note = {clmc},
  crossref = {p10519},
  url = {http://clmc.usc.edu/publications/K/kalakrishnan-ICML2012.pdf}