Active sensorimotor control for tactile exploration

In this paper, we present a novel and robust Bayesian approach for autonomous active exploration of unknown objects using tactile perception and sensorimotor control. Despite recent advances in tactile sensing, robust active exploration remains a challenging problem, which is a major hurdle to the practical deployment of tactile sensors in robots. Our proposed approach is based on a Bayesian perception method that actively controls the sensor with local small repositioning movements to reduce perception uncertainty, followed by explorative movements based on the outcome of each perceptual decision making step. Two sensorimotor control strategies are proposed for improving the accuracy and speed of the active exploration that weight the evidence from previous exploratory steps through either a weighted prior or weighted posterior. The methods are validated both off-line and in real-time on a contour following exploratory procedure. Results clearly demonstrate improvements in both accuracy and exploration time when using the proposed active methods compared to passive perception. Our work demonstrates that active perception has the potential to enable robots to perform robust autonomous tactile exploration in natural environments.