Peg-in-Hole assembly under uncertain pose estimation

A large part of industrial assembly tasks can be broken down to permutations of Peg-in-Hole actions. In this paper we compare the success rate of such actions using traditional robot control to an adaptive control strategy based on Dynamic Motor Primitives. Both methods use vision-based pose estimation to locate the peg and the hole in the workspace of an industrial robot, and a dexterous gripper to grasp the peg. The experiments show that the traditional robot control performs the PiH action with a low success rate, while the DMP based PiH can achieve a high success rate.