Adaptive human-inspired compliant contact primitives to perform surface–surface contact under uncertainty

Abstract

This paper focuses on devising a control policy that is inspired by human strategy to enable robots to perform surface–surface contact between a hand-held object (e.g. a box) and the environment (e.g. table). We assume the object’s shape is partially known and consider uncertainties in both the environment and the object grasp. Our analysis on ten untrained subjects indicates that during this task: (1) the subjects start decreasing the angular velocity before the complete alignment (in contrast to existing approaches), and (2) they do not control the contact force to remain at a fixed value. Our study is also consistent with the hypothesis that the subjects determine an over-estimate of the relative angle between the object in hand and the environment. Based on these observations we propose a novel control policy, called Surface-Surface Contact Primitive (SSCP), which can perform the surface–surface alignment task with only partial information about the hand-held object and the environment. Furthermore, SSCP only requires a rough estimate of the surface normal vector and does not rely on either the estimation of contact type (i.e. point or edge contact) or locations of contact points. We evaluate the performance of the proposed controller on a set of robot experiments using two seven-degree-of-freedom robots, one for imposing uncertainty to the environment, and the other to perform the experiment. We show the applicability of the controller on four objects with different geometries, its generalization to four different surface materials, and its robustness to uncertainty in environment and grasp as well as external perturbations.