MPPI-DWIT: Robust Bearing-Only Target Tracking of Single AUV With Detection Sector Constraints

Continuous underwater bearing-only target tracking with limited sensor detection sectors presents a significant challenge for Autonomous Underwater Vehicles (AUVs). This challenge arises from the tight coupling between target state estimation and AUV trajectory optimization. This paper proposes the Model Predictive Path Integral with Dynamic Weight Integration Tracking (MPPI-DWIT) framework to solve the problem of bearing-only target tracking under limited sensor detection sectors, by integrating an observability metric based on Fisher Information Matrix (FIM) into the MPPI path planner. The observability metric directly improves the accuracy of target state estimation during trajectory sampling. The framework uses a hard-boundary strategy to strictly enforce sensor detection sectors limits and AUV maneuverability constraints. An innovative dynamic weight integration mechanism is proposed, which automatically balances the need to maximize observability with the need to maintain a safe tracking distance. Extensive Monte Carlo simulations show that MPPI-DWIT outperforms the state-of-the-art methods, in terms of estimation accuracy, tracking times, and failure rates. Real-world single AUV tracking trials further confirm the effectiveness and robustness of MPPI-DWIT.