Model Predictive Control for Motion Planning of Quadrupedal Locomotion

This paper is motivated to transfer the model predictive control approach used in bipedal locomotion to formulate gait planning of quadruped robots. The particular lateral-sequence gait of quadrupeds is treated as an equivalence to the bipedal walking. The model predictive control algorithm uses 3D-linear inverted pendulum model for representing the center of mass dynamics for planning the quadrupedal gaits, and a dimensionless discrete-time state-space formulated is derived for model predictive control. Subsequently, the footholds can be generated automatically via optimization of quadratic programming without the need of a separate footstep planner. The generated walking gaits were implemented and validated first in the physics simulation of a quadruped named EHbot, and then the effectiveness of the proposed method was further demonstrated through our experiments. Both simulation and experimental data are presented and analyzed for evaluating the performance.