Intelligent and environment-independent Peg-In-Hole search strategies

Abstract

Cylindrical Peg-In-Hole assembly has been the benchmark force-controlled robotic assembly. It involves two main stages. The first one aims at placing the peg center within the clearance region of the hole center, known as the search phase. The next step is to correct the orientational misalignment, known as the insertion phase. The insertion has been widely researched as compared to the search phase. Search is generally done by locating the hole center using a vision-sensor or by using blind search techniques. An intelligent search in which a neural network is trained with the moment-profile over the hole surface and then tries to infer the hole center based on the current moment values, works for a parallel peg only and also requires the moment profile over the complete hole and hence is not general. This paper generalizes this approach for the tilted peg case. Another intelligent strategy is precession-based hole search, but again, it requires the tilt of the peg and peg's center to be known to perform precession. We need a general-purpose intelligent search strategy that can work without any knowledge of the environment (like hole diameter, peg tilt, etc.). In this paper, we suggest search strategies using mathematical optimization techniques. These strategies do not require any a priori information about the working environment. Simulation results are presented.