Contact Impedance Adaptation via Environment Identification

Abstract :

In this paper we present the results of an approach for identifying the environment using Bayesian inference methods. Using this information, the contact properties between a robotic manipulator and a particular scenario are regulated by means of an impedance controller that adapts to the identified environment. Off-line, the robot records sensory data from a set of possible environments and computes their likelihood functions to be used in a Bayesian estimation model. Online, the robot contacts an environment, computes the posterior probabilities using Bayes’ rules, and determines the environment with highest confidence. This information modifies the behaviour of an impedance controller that regulates the robot-environment contact interaction. Simulation and experimental results with an industrial robotic manipulator (Mitsubishi PA-10) are shown that depict the performance of the presented approach.