Contact Localization using Velocity Constraints

Localizing contacts and collisions is an important aspect of failure detection and recovery for robots and can aid perception and exploration of the environment. Contrary to state-of-the-art methods that rely on forces and torques measured on the robot, this paper proposes a kinematic method for proprioceptive contact localization on compliant robots using velocity measurements. The method is validated on two planar robots, the quadrupedal Minitaur and the two-fingered Direct Drive (DD) Hand which are compliant due to inherent transparency from direct drive actuation. Comparisons to other state-of-the-art proprioceptive methods are shown in simulation. Preliminary results on further extensions to complex geometry (through numerical methods) and spatial robots (with a particle filter) are discussed.

I. Introduction

Robots are proving to be increasingly useful in unstructured environments, such as cluttered homes and outdoor terrain. However, in these environments robots must deal extensively with the making and breaking of contact with uncertain or unknown object shapes and poses. This makes contact localization a vital skill. For example, Fig. 1 shows a legged robot walking up stairs, with one of its legs in contact with the edge of a step. In this scenario, noisy estimation of the height of a stair can lead to unexpected contact between the edge and the leg causing the robot to trip. Another example is an industrial robot gripper with planar fingers approaching an object to grasp or estimate its surface, Fig. 9.