Robust discovering and tracking in challenging environments

While wireless sensor network aided robots lend humans a hand in accessing to certain difficult and/or otherwise unreachable areas, the robots may face challenges in locomotion, object discovering, searching and tracking among many demanding tasks. The unreliable wireless communication and possible components failures of wireless sensor nodes coupling with the noises in the uncertain environments further complicate the situation. To address this issue, we develop a robustified estimation scheme that's capable of online rectifying outliers and detecting anomalies including sensor faults. By integrating a recursive variant of the M-estimator into the Kalman filter via an recursively reweighted least squares implementation, it robustifies the Kalman filter's performance upon outliers without scarifying the optimality of the latter. Moreover, we employ a General Likelihood Ratio (GLR) test to further fine tune the detection of changes in the environment and wireless sensor nodes, where the false alarm constraint is achieved through Monte Carlo simulation. The effectiveness of the idea is illustrated through experiments and synthetic data at the current stage.