Autonomous Navigation Method Based on Deep Reinforcement Learning with Dual-Layer Perception Mechanism

Autonomous navigation systems are crucial in the field o f r obotics. Traditional methods o ften require extensive manual parameter tuning, which is time-consuming. In this paper, we present an autonomous navigation method that leverages deep reinforcement learning (DRL) enhanced by a dual-layer perception mechanism. This method integrates raw sensor data, local grid map, and the goal pose as inputs. It utilizes the DRL framework to develop autonomous navigation strategies and directly generates robot action commands. This method eliminates the need for manual parameter adjustment, relying solely on continuous trial-and-error training to enable autonomous navigation. Comparative experiments in a simulation environment reveal that this system offers enhanced robustness and scalability compared to other DRL-based autonomous navigation systems.