Certifiable cooperative localization for underwater navigation

Abstract

Accurate underwater positioning remains one of the most significant obstacles to autonomous underwater vehicle (AUV) operations. Satellite-based navigation signals are unavailable underwater, so AUVs must dead-reckon using inertial sensors, coupled with velocity or heading references. Due to random noise and variable biases in inertial sensor measurements, the AUV’s position uncertainty grows steadily over the course of the mission, but can be reduced through range measurements to fixed or mobile references. The associated range-aided simultaneous localization and mapping (SLAM) problem is particularly challenging to solve with existing optimization methods. Individual range measurements provide limited geometric constrains on vehicle position and are subject to non-linear errors due to multi-path propagation. Attempts to optimize typical range-aided SLAM cost functions often return solutions which represent local, rather than global minima, resulting in unpredictable vehicle behavior when used for closed-loop navigation. This thesis applies a recently developed certifiable optimization algorithm, Certifiably Correct Range-aided SLAM (CORA), to the problem of cooperative localization between AUVs. CORA leverages aspects of the range-aided SLAM problem structure to find solutions which can be certified to be globally optimal. This method is integrated into a novel cooperative localization scheme, in which each vehicle maintains a locally held, periodically updated copy of the centralized, multi-agent factor graph. The cooperative localization framework presented here leverages acoustic modems for both range measurement and the sharing of sub-graphs through inter-vehicle communication. This approach was validated through extensive field trials using two modular, low-cost Spurdog AUVs were equipped with WHOI Micromodem2 payloads. Results from single and multi-vehicle deployments demonstrated that CORA substantially outperforms existing solvers when faced with poor landmark initialization and reduced observability as a result of real-world communication failures. The results presented here demonstrate the added value of coupling certifiable estimation with cooperative localization for multi-AUV localization problems, particularly in challenging, GPS-denied environments.