Contact-Invariant Optimization for Hand Manipulation

Abstract

We present a method for automatic synthesis of dexterous hand movements, given only high-level goals specifying what should happen to the object being manipulated. Results are presented on a wide range of tasks including grasping and picking up objects, twirling them between the fingers, tossing and catching, drawing. This work is an extension of the recent contact-invariant optimization (CIO) method, which introduced auxiliary decision variables directly specifying when and where contacts should occur, and optimized these variables together with the movement trajectory. Our contribution here is extending the unique contact model used in CIO which was specific to locomotion tasks, as well as applying the extended method systematically to hand manipulation.