Active semantic localization of mobile robot using partial observable Monte Carlo Planning

This paper proposes a new active localization approach based on the semantic map. The deterministic active localization problem is modeled in POMDP (Partial Observable Markov Decision Process) framework and solved using POMCP (Partial Observable Monte Carlo Planning algorithm). The new approach provides a general heuristic search which outperforms the traditional greedy strategy based techniques in active localization. To provide better heuristic, a mixed reward function is defined, which combines uniqueness of observation and entropy reduction, and shows a good performance in the simulation experiments.