LEARNING DEXTEROUS MANIPULATION SKILLS FOR MULTIFINGERED ROBOT HANDS USING THE EVOLUTION STRATEGY

We present a method for autonomous learning of dextrous manipulation s kills with multifingered robot hands. We use heuristics derived from o bservations made on human hands to reduce the degrees of freedom of th e task and make learning tractable. Our approach consists of learning and storing a few basic manipulation primitives for a few prototypical objects and then using an associative memory to obtain the required p arameters for new objects and/or manipulations. The parameter space of the robot is searched using a modified version of the evolution strat egy, which is robust to the noise normally present in real-world compl ex robotic tasks. Given the difficulty of modeling and simulating accu rately the interactions of multiple fingers and an object, and to ensu re that the learned skills are applicable in the real world, our syste m does not rely on simulation; all the experimentation is performed by a physical robot, in this case the 16-degree-of-freedom Utah/MIT hand . Experimental results show that accurate dextrous manipulation skills can be learned by the robot in a short period of time. We also show t he application of the learned primitives to perform an assembly task a nd how the primitives generalize to objects that are different from th ose used during the learning phase.