A Particle-Filter Approach for Active Perception in Networked Robot Systems

The presence of children in a social assistive robotics context is particularly challenging for perception, mainly, in the task of locating them using inherently uncertain sensor data. This paper proposes a method for active perception with the goal of finding one target, e.g., a child wearing a RFID tag. This method is based on a particle-filter modeling a probability distribution of the position of the child. Negative measurements are used to update this probability distribution and an information-theoretic approach to determine optimal robot trajectories that maximize information gain while surveying the environment. We present preliminary results, in a real robot, to evaluate the approach.