A Synthesized Neural Control System for Bioinspired Robots to Achieve Diverse Locomotion

Abstract

There is great potential for legged robots in unstructured environments. However, model‐based approaches benefit from precise model analysis, which can be cumbersome and demand substantial domain expertise, while learning‐based methods, though promising, often necessitate prolonged training periods and may result in complex and opaque controllers. This architecture aims to mimic neural‐muscle control and sensory feedback mechanisms, enabling legged robots to adjust neural signal intensity based on proprioceptive feedback and achieve behavior responses similar to those observed in animals. Specifically, and the central pattern generator creates insect‐like gaits, the virtual motoneurons network generates continuously adjustable trajectories for omnidirectional motion and limb control. The sensorimotor integration module, event‐based finite state machine, and local reactive strategies allow robots to traverse unstructured terrains. The method is experimentally applied to a newly developed hexapod robot named RENS H2. The results indicate that the proposed method enhances the robot's locomotion diversity, enabling adaptive navigation in unstructured terrains, including overcoming steps with heights up to 66.7% of its leg length.