一种室内未知环境下的无人车分层自主导航方法

To enhance the autonomous navigation ability of unmanned vehicles in typical indoor unknown environments, a hierarchical autonomous navigation method is proposed, which realizes autonomous navigation without relying on maps by sensing the environment through on-board sensors. In the global strategy layer, Gaussian process based frontier detection is combined with corridor end vanishing point detection to supply the unmanned vehicle with essential global information, thereby improving navigation efficiency in typical indoor settings. In the local planning layer, a self attention mechanism is integrated into a proximal policy optimization framework and the reward function is refined. These enhancements improve the algorithm’s ability to extract environmental features and ensure continuity of vehicle actions. Furthermore, a sub-goal selection strategy is devised to decompose the complex navigation task into multiple smaller segments, allowing the vehicle to advance toward the overall goal in stages. Simulation results demonstrate that the proposed method can complete navigation tasks both efficiently and stably. Compared with the mature methods based on map frontiers, the average path length is reduced by 11.92% and the average runtime is shortened by 24.94%.