Belief Driven Manipulator Control for Integrated Searching and Tracking

This paper presents a feedforward control strategy for a robotic manipulator based on a belief function. The belief about a target's next location, as described by a probability density function, is maintained by a recursive Bayesian process that fuses observations with a target motion model. A sensor model that incorporates positive and negative sensor readings allows the single belief function to be used to deliver both searching and tracking behaviors. Constrained non-linear optimization is used to search configuration space for the control action that maximizes the subsequent probability of detection. To demonstrate application of the technique, a simple example is elaborated for a searching and tracking task with an eye-in-hand sensor