Integrating Contact-Aware CPG System for Learning-Based Soft Snake Robot Locomotion Controllers

Xuan Liu, Cagdas D. Onal, Jie Fu

Abstract: Contact-awareness poses a significant challenge in the locomotion control of soft snake robots. This article is to develop bioinspired contact-aware locomotion controllers, grounded in a novel theory pertaining to the feedback mechanism of the Matsuoka oscillator. This mechanism enables the Matsuoka central pattern generator (CPG) system to function analogously to a “spinal cord” in the entire contact-aware control framework. Specifically, it concurrently integrates stimuli, such as tonic input signals originating from the “brain” (a goal-tracking locomotion controller) and sensory feedback signals from the “reflex arc” (the contact reactive controller), for generating different types of rhythmic signals to orchestrate the movement of the soft snake robot traversing through densely populated obstacles and even narrow aisles. Within the “reflex arc” design, we have designed two distinct types of contact reactive controllers: 1) a reinforcement learning-based sensor regulator that learns to modulate the sensory feedback inputs of the CPG system, and 2) a local reflexive controller that establishes a direct connection between sensor readings and the CPG's feedback inputs, adhering to a specific topological configuration. These two reactive controllers, when combined with the goal-tracking locomotion controller and the Matsuoka CPG system, facilitate the implementation of two contact-aware locomotion control schemes. Both control schemes have been rigorous tested and evaluated in both simulated and real-world soft snake robots, demonstrating commendable performance in contact-aware locomotion tasks. These experimental outcomes further validate the benefits of the modified Matsuoka CPG system, augmented by a novel sensory feedback mechanism, for the design of bioinspired robot controllers.