Bending Perception-Based Variable Stiffness Control for Snake Robots in Pipe Navigation
Shiyong Meng,, Huizhuo Yang, Kai Shen, Honglu Xu, Gen Chen, Jianming Wang,, and Xuan Xiao∗
AI summary
Problem
Snake robots frequently jam or waste energy in complex pipelines due to fixed compliance parameters and a lack of real-time environmental perception without relying on external sensors or human guidance.
Approach
The method uses an LSTM network to classify pipe bend angles from the first five joint angles, then propagates a Gaussian-shaped stiffness wave from head to tail to locally soften the robot at bends while maintaining propulsion stiffness elsewhere.
Key results
- LSTM bend detection achieves over 93% accuracy using only joint angle data
- Variable stiffness wave reduces energy consumption and prevents jamming at sharp bends
- Successfully traversed a 5-meter complex pipeline with an 83% valid trial success rate
- Optimized stiffness parameters balance traversal speed (~3.7 cm/s) and energy efficiency across different curvatures
Why it matters
Offers a practical, sensorless adaptive control strategy that improves the reliability and energy efficiency of snake robots for industrial pipe inspection and maintenance.
Abstract
In this paper, Bending Perception-Based Variable Stiffness Control (BP-VSC) is proposed, which enables the snake robot to efficiently navigate complex pipelines. First, the method includes an LSTM-based pipe angle detection model and a control method that transmits a stiffness wave from head to tail. Second, we design a cable-free distributed experimental platform to prevent restrictions caused by cables during pipeline navigation. Finally, a series of experiments are conducted. We optimize the control parameters of BP- VSC in the constructed pipe and evaluated the method. The results demonstrate that BP-VSC significantly enhances the performance of the snake robot in navigating complex pipelines.