The linear inverted pendulum (LIP) model with preview control is a standard approach for gait planning in bipedal humanoid robots. However, its inherent simplifications lead to significant model errors that compromise walking stability. To address this, we propose a compensation strategy that integrates two core components: a compensation controller based on a multi-link model and a joint torque trajectory generator based on full robot dynamics. Extensive simulations in ROS/MuJoCo demonstrate that under the same reference zero moment point (ZMP) trajectory, the root mean square error (RMSE) of ZMP tracking is reduced by 50.19% (from 0.5212 m to 0.2596 m). The proposed approach thus provides a significant enhancement to the LIP framework, improving the walking stability and tracking accuracy of bipedal robots.
Wang et al. (Fri,) studied this question.