Snake Robot with Tactile Perception Navigates on Large-scale Challenging Terrain

Abstract: Along with the advancement of robot skin technology, there has been notable progress in the development of snake robots featuring body-surface tactile perception. In this study, we proposed a locomotion control framework for snake robots that integrates tactile perception to augment their adaptability to various terrains. Our approach embraces a hierarchical reinforcement learning (HRL) architecture, wherein the high-level orchestrates global navigation strategies while the low-level uses curriculum learning for local navigation maneuvers. Due to the significant computational demands of collision detection in whole-body tactile sensing, the efficiency of the simulator is severely compromised. Thus a distributed training pattern to mitigate the efficiency reduction was adopted. We evaluated the navigation performance of the snake robot in complex large-scale cave exploration with challenging terrains to exhibit improvements in motion efficiency, evidencing the efficacy of tactile perception in terrain-adaptive locomotion of snake robots.

• The paper introduces a hierarchical reinforcement learning strategy that integrates tactile perception to improve snake robot adaptability on challenging terrains.
• Leveraging a two-phase curriculum and distributed RL, the method efficiently trains tactile-adaptive controllers for precise gait selection.
• Simulation tests in random cave environments demonstrate superior navigation performance, paving the way for advanced search-and-rescue and exploration applications.

Introduction

The development of robotics has steered toward creating machines capable of navigating challenging terrains, reminiscent of natural creatures. Among these, snake robots have gained prominence due to their flexibility and ability to traverse narrow spaces. These robots consist of multiple interconnected joints, enabling them to perform complex locomotion patterns. Common control strategies have utilized dynamic modeling, feedback control, Central Pattern Generators (CPGs), and serpenoid curves to govern these movements.

Tactile Adaptation and Reinforcement Learning

The integration of tactile sensors on snake robots presents a novel capability: whole-body tactile perception. This allows the robots to detect contact forces at various points along their bodies, providing them with an intricate understanding of surface characteristics like roughness or slope. Recognizing the potential of tactile feedback, this study introduces a hierarchical reinforcement learning (HRL) architecture to create a control system that enhances the adaptability of snake robots to difficult terrains such as large-scale caves.

Hierarchical Control Scheme and Training

The architecture divides the navigation challenge into three levels. At the highest level, global navigation involves path planning and waypoint generation. On the local navigation level, reinforcement learning (RL) governs how the robot adapts to immediate terrain using tactile feedback to steer from waypoint to waypoint. Finally, the lowest level relies on practical joint controllers to carry out the movement commands.

To train the local navigation control system, a two-phase curriculum learning strategy was employed. Initially, basic gaits are learned without tactile inputs. Following this, a tactile-adaptive local navigation control scheme uses RL models called Adaptors, which take tactile information from adjacent body parts and make real-time decisions on gait selection. This second phase ensures the snake robot develops the ability to adapt its locomotion patterns based on physical sensations, learning to traverse previously unseen terrains.

Simulation and Results

The simulation phase highlighted the challenges of using tactile sensors for RL due to computational intensity. To address this, a distributed reinforcement learning framework was used to distribute the computational load across multiple machines, significantly speeding up the training process. Extensive testing within randomly generated cave environments showed that the snakes with tactile-adaptive control outperformed traditional RL controllers, demonstrating superior navigation ability and increased robustness when encountering new terrains.

Conclusion

The study's findings confirm that tactile perception can significantly improve the terrain adaptability of snake robots. By combining hierarchical reinforcement learning with a curriculum structure and distributed computing, this approach paves the way for flexible robots capable of performing versatile and advanced tasks in search-and-rescue operations, planetary exploration, and environments inaccessible to humans and traditional robots. Future research is anticipated to undertake real-world testing and further refine this inventive control strategy.