Tactile Correction and Multiple Training Data Sources for Robot Motion Control
This work considers our approach to robot motion control learning from the standpoint of multiple data sources. Our paradigm derives data from human teachers providing task demonstrations and tactile corrections for policy refinement and reuse. We contribute a novel formalization for this data, and identify future directions for the algorithm to reason explicitly about differences in data source.
Record created on 2010-01-19, modified on 2016-08-08