Adaptive Learning for Controlled Lagrangian Particle Tracking

Controlled Lagrangian particle tracking (CLPT) is a feedback control methodology for autonomous underwater vehicles (AUVs) in dynamic ocean environments that can reveal information about ocean flow and model error through the operation of AUVs. From CLPT, we evaluate the navigational performance of AUVs with controlled Lagrangian prediction error (CLPE), which represents the difference between the predicted and true trajectories of AUVs. In this paper, using CLPE, we develop an adaptive learning algorithm that allows AUVs to learn their true motion in ocean flow fields in real time. While previous work shows that CLPE increases over time in simulated and field experiments, the proposed algorithm forces CLPE to converge to zero so that the vehicle motion model is unique and flow fields can be identified. For the robustness of the proposed algorithm, we prove that CLPE is ultimately bounded under disturbances. The proposed algorithm is verified by simulation results.