Adaptive Informative Path Planning for Lightweight AUV in Coastal Benthic Environments

Efficient coastal benthic monitoring requires au-tonomous underwater vehicles (AUVs) that can systematically survey areas while adaptively investigating features of interest. Traditional coverage planning methods follow fixed patterns that may miss important targets or waste resources on less informative regions. This paper presents a reinforcement learning approach to adaptive informative path planning (AIPP) that unifies coverage completion with target detection objectives. Our framework combines a carefully structured reward mechanism with a novel local view feature extraction method to enable ef-ficient exploration and target investigation. Experimental results demonstrate that our approach achieves 97.8 % coverage rate and 87.9% detection efficiency across 1000 simulated test scenarios, with a 97.3% task success rate. The framework shows particular strength in environments with patchy target distributions, typical of coastal benthic ecosystems. While simulation results are promising, we identify key challenges in parameter optimization and uncertainty handling that must be addressed for real-world deployment. This work establishes a foundation for more efficient autonomous benthic observation systems that can effectively balance systematic coverage with adaptive information gathering.