Snake robots exhibit significant potential in complex terrain exploration due to their exceptional flexibility and environmental adaptability. However, conventional vision-based navigation approaches demonstrate substantial limitations in low-light or completely dark environments (e.g., underground pipelines, disaster ruins), requiring reliance on high-power illumination systems or costly infrared/radar sensors, which lead to substantial increases in system costs and energy consumption. Furthermore, multi-sensor integration in confined spaces introduces structural complexity and reliability challenges. This paper proposes a vision-independent autonomous control method for serpentine robots based on reinforcement learning. The approach establishes a simulation environment incorporating walls, obstacles, and low-clearance passages, while implementing multi-faceted reward functions and optimized training strategies to achieve vision-free navigation in complex terrains. Leveraging the Proximal Policy Optimization (PPO) algorithm with joint state and positional information, the proposed method enables visual-input-free autonomous navigation through low-clearance scenarios. Experimental results demonstrate enhanced locomotion stability and improved navigational efficacy in confined spaces within simulated environments. This work provides a novel solution for vision-independent robotic control systems, particularly addressing challenges in unstructured subterranean environments.