Towards experimental verification of an automated compliant motion planner based on a geometric theory of error detection and recovery

In earlier work, B.R. Donald (1988, 1989) implemented an automated compliant motion planner based on a geometric theory of error detection and recovery (EDR). In the present work, an attempt is made to validate the theory, in part, by executing the plans it has generated on a physical force-controlled robot. The planner, LIMITED, accepts as input geometric descriptions of the environment and bounds on sensing and control uncertainties, and synthesizes motion sequences that, when executed, recognizably achieve the goal when possible and signal failure otherwise. The motions are compliant; that is, surfaces can be used to guide motions toward a goal. Experimental results attained in executing two plans earlier generated by LIMITED are presented. The first plan is for the task of inserting a planar peg into a planar hole, with geometric model uncertainty. This plan was found to be quite robust. The second plan meshes two gears in the plane. This plan proved less robust than the peg insertion plan, and the authors examine why this is so