Reducing uncertainty in robotic surface assembly tasks based on contact information

In contrast to hard automation, flexible assembly using a general-purpose robot holds the promise of reducing costs and increasing productivity of small batch products as well as adaptivity to product changes. However, robotic assembly remains one of the most challenging problems in robotics research. In particular, it has to overcome uncertainty in the environment and the location of parts. This paper addresses how to estimate and correct part positioning uncertainties with respect to the assembly task by using the contact trajectory data collected during active part interaction. We first devise a nominal assembly motion sequence based on an approximate knowledge of the poses of the parts, which includes exploratory compliant moves for the purpose of parameter estimation. As the robot manipulator executes the nominal motion sequence, data collected during exploratory compliant moves are used to estimate part poses more accurately, and the improved information is used to update the subsequent assembly motion sequence to correct errors in the nominal plan. We implemented the strategy using a back-drivable Barrett WAM robotic arm without a force sensor and BarrettHand BH8-262. The empirical results have demonstrated the effectiveness of our approach.