Multirobot Belief Fusion for Mapping and Tracking with Unreliable Sensors

We present an approach for mapping and tracking with multiple sensors that is robust to false-positives and falsenegatives. Sequential observations are first passed through the Improbability Filter to exclude false-positives, combined into a single joint observation, and finally fed to the Kalman filter to update the prior joint belief. Modifications to the Kalman update equations eliminate biases introduced by using both filters in succession; similar dual-filter techniques ignore bias or address it in an ad-hoc, unprincipled fashion. We apply the framework to the domain of robot soccer; in simulation, our algorithm, which we call the Comprehensive Kalman filter or CKF, far outperforms the standard Kalman filter when applied to landmark mapping in the presence of false-positives and false-negatives. CKF achieves convergence even with sensors of false-positive rates up to 50%.