Extremely Fast Computation of CoM Trajectory Generation for Walking Leveraging MPPI Algorithm

Abstract

This study presents a novel approach for a highly efficient solver that can reduce the computational burden for the center of mass (CoM) trajectory in robot walking. According to the heavy computation costs, conventional trajectory generation methods pose challenges when it comes to real-time trajectory generation or their implementation on humanoid robots with low-cost boards. Our approach enables fast computation by an efficient sampling method within a model predictive path integral framework. As a result, the linear inverted pendulum model-based CoM trajectory can be achieved by an average computation time of 0.09 ms with a sample time of 0.5 ms. The performance and effectiveness of the proposed method are verified through numerical comparisons and humanoid robot walking simulations.