Proactive Body Joint Adaptation for Energy-Efficient Locomotion of Bio-Inspired Multi-Segmented Robots

Abstract

Typically, control strategies for legged robots have been developed to adapt their leg movements to deal with complex terrain. When the legs are extended in search of ground contact to support the robot body, this can result in the center of gravity (CoG) being raised higher from the ground and can lead to unstable locomotion if it deviates from the support polygon. An alternative approach is body adaptation, inspired by millipede/centipede locomotion behavior, which can result in low ground clearance and stable locomotion. In this study, we propose novel proactive neural control with online unsupervised learning, allowing multi-segmented, legged robots to proactively adapt their body to follow the surface contour and maintain efficient ground contact. Our approach requires neither kinematics nor environmental models. It relies solely on proprioceptive sensory feedback and short-term memory, enabling the robots to deal with complex 3D terrains. In comparison to traditional reflex-based control, our approach results in smoother and more energy-efficient robot locomotion on terrains with concave and convex curves or slopes of varying degrees in both simulation and real-world implementation.