Proprioceptive Localization for Mobile Manipulators

Abstract

We use a combination of laser data, measurements of joint angles and torques, and stall information to improve localization on a household robotic platform. Our system executes trajectories to collide with its environment and performs probabilistic updates on a distribution of possible robot positions, ordinarily provided by a laser range finder. We find encouraging results both in simulations and in a real-world kitchen environment. Our analysis also suggests further steps in localization through proprioception.