Informative Path Planning for AUV-based Underwater Terrain Exploration with a POMDP

Autonomous underwater vehicles (AUVs) in underwater terrain exploration applications represent a topic area, and the interesting problem is planning paths to maximize the vehicle information gathered and combining this information to build a complete map. The Gaussian process (GP) is utilized as a basic environment model and updated using the Bayesian data fusion technique with sensor information. A path planning algorithm, which formulates the terrain exploration problem as the finite-horizon partially observable Markov decision process (POMDP), is proposed to overcome the limitation of the planner converge to locally suboptimal solutions. In addition, a Monte Carlo Tree Search based on the motion primitives tree (MPT-MCTS) solver is developed to solve this POMDP. The effectiveness of the proposed method is explored in the simulation experiment, and its potential is demonstrated by comparing it with other optimization algorithms.