Bipedal locomotion is a complex field in both biology and robotics. As modern AI-enhanced control methods continue struggling to manage compliant structures, artificial bipedal locomotion still lacks the natural agility and adaptability of biological systems. To bridge this gap, a novel natural control paradigm is developed, inspired by the human gait and neuromuscular complex. The approach employs antagonistic muscular control as a low-level hybrid impedance control strategy for series elastic actuators. A probabilistic behavior architecture provides an efficient high-level control framework. Both enable natural adaptability and resilience against disturbed motions in uncertain environments. Experimental results on the bipedal robot CARL demonstrate the effectiveness of the natural control system in replicating human-like control characteristics.
Patrick Vonwirth (Thu,) studied this question.