Collision-free active sensing for maximum seeking of unknown environment fields with Gaussian processes

Abstract

This paper presents a collision-free active sensing algorithm that safely and efficiently searches for the max-imum point while reconstructing the unknown environment field. Bayesian optimization (BO) for optimizing the unknown function with Gaussian processes (GPs) is used for active sensing with a new acquisition function. Besides, the mobile sensor estimates Euclidean signed distance field using GPs to avoid obstacles with its fast collision checking capability. To mitigate the local maximum problem, Monte Carlo tree search (MCTS), one of state-of-the-art planning techniques, is adopted as a non-myopic planner. In particular, obstacle avoidance and active sensing are integrated into a unified framework using a safe BO algorithm (known as SafeOpt-MC) based on GPs and MCTS. Numerical simulations are performed to validate the feasibility and performance of the proposed framework with a diverse set of environments.