Deep Reinforcement Learning-Based Navigation with RGB-D Input for Autonomous Systems

This paper presents a novel navigation framework based on deep reinforcement learning (DRL) and dual-source network architecture, it utilizes RGB-D input for autonomous mobile robots in complex environments. The integration of RGB-D data into the dual-source Actor-Critic architecture allows enhanced spatial awareness, obstacle avoidance, and path planning, addressing limitations in RGB-only and monocular vision systems. Experimental results demonstrate significant improvements in navigation accuracy, robustness, and safety in both simulated and real-world scenarios. Our approach advances state-of-the-art in DRL-based autonomous navigation by fully leveraging RGB-D data for complex navigation tasks.