Adaptive Neuromechanical Control for Robust Behaviors of Bio-Inspired Walking Robots

Walking animals show impressive locomotion. They can also online adapt their joint compliance to deal with unexpected perturbation for their robust locomotion. To emulate such ability for walking robots, we propose here adaptive neuromechanical control. It consists of two main components: Modular neural locomotion control and online adaptive compliance control. While the modular neural control based on a central pattern generator can generate basic locomotion, the online adaptive compliance control can perform online adaptation for joint compliance. The control approach was applied to a dung beetle-like robot called ALPHA. We tested the control performance on the real robot under different conditions, including impact force absorption when dropping the robot from a certain height, payload compensation during standing, and disturbance rejection during walking. We also compared our online adaptive compliance control with conventional non-adaptive one. Experimental results show that our control approach allows the robot to effectively deal with all these unexpected conditions by adapting its joint compliance online.