Loss of leg function in quadruped robots often leads to severe mobility degradation in complex environments. Inspired by the effective forelimb utilization of the oriental mole cricket, this paper proposes a cooperative compensation strategy that integrates morphological design, gait optimization, and reinforcement learning to address single-leg failure. A novel configuration of a sprawled-posture quadruped robot equipped with two functional arms is developed. Based on kinematic analysis and the longitudinal stability margin criterion, a dynamic gait adjustment method is derived. Two control frameworks—a Central Pattern Generator (CPG) model and a Deep Deterministic Policy Gradient (DDPG) reinforcement learning approach—are developed and evaluated. Simulation results demonstrate that the bio-inspired arm compensation effectively mitigates yaw motion during foreleg failure. For the more challenging hindleg failure, the reinforcement learning controller significantly outperforms CPG limiting yaw fluctuation to a narrow range of -0.11 to -0.09 rad. Physical experiments validate that the proposed strategy restores walking performance to over 85% of the normal performance. After reinforcement learning training, mobility is not only fully recovered but slightly surpasses normal performance, with yaw error approaching zero. This work demonstrates a "morphology–control" cooperative compensation paradigm, enhancing fault tolerance and robustness of legged robots in unstructured hazardous environments.
Guo et al. (Sun,) studied this question.