This paper presents a motion control algorithm that exploits mutual information and a Bayesian filter to optimally guide a mobile robotic sensor, e.g., an unmanned aerial or ground vehicle (UAV or UGV) with a sensor, to localize an unknown target such as the source of a gas/chemical leak. Specifically, optimal feedforward inputs are found such that with respect to the posterior distribution, the robot moves to minimize uncertainty. The formulation depends on the robot’s dynamics model and the sensor’s stochastic measurement model. Additionally, a utility function is defined such that the estimator’s uncertainty is minimized, i.e., the acquisition of information is maximized. The approach is applied to a single robot with three different sensor models for validation. In particular, for the chemical concentration sensor case a Gaussian plume likelihood model is assumed and simulation results show that a single robot can effectively localize the unknown source, demonstrating the effectiveness of the approach.

This content is only available via PDF.
You do not currently have access to this content.