A Vision Based Architecture for Continuous Path Decision Making with DDPG

Automated guided vehicle (AGV) systems are often equipped with multiple sensors to ensure accurate navigation, but such configurations increase both cost and implementation complexity. For smaller-scale applications or structured environments such as warehouses and factory aisles, a lightweight solution based on monocular vision can offer a more practical alternative, though it remains sensitive to illumination changes and lacks depth perception. This paper introduces VINA (Vision-based Intelligent Navigation Architecture), a framework that applies the Deep Deterministic Policy Gradient (DDPG) algorithm to achieve vision-based autonomous navigation using a single camera. Experimental results show that the system can perform directional prediction with an accuracy of 87.46%. While the results confirm the feasibility of the approach, the simplicity of the test track limits the extent to which the advantages of reinforcement learning can be fully observed. Future work will therefore extend VINA to more complex and dynamic environments to better evaluate its scalability and robustness.