Integration Framework for Offline Trajectory Optimization and Online Model Predictive Control for Legged Robots

Abstract

In the last decade, legged mobile robots have gained notoriety for their ability to move safely over rough terrain and overcome obstacles such as slopes and stairs, opening up new applications compared to wheeled mobile robots. New developments that improve the robustness of trajectory planning and dynamic control of legged robots are crucial for the advancement of this field. The aim of this work is to develop a framework that integrates offline trajectory optimization for legged robots with online Model Predictive Control (MPC) while taking into account the elevation map of the terrain. The trajectory optimization is based on the open-source library TOWR (Trajectory Optimization for Walking Robots), which employs a continuous function to represent the map of the terrain. To make it more generic, an interface was implemented to allow 2.5D elevation maps to be used as terrain representation. Furthermore, the trajectories generated by TOWR are provided as references for a MPC implemented based on the open-source library OCS2. The trajectories optimized by the MPC are then tracked by a weighted Whole-Body Controller (WBC), which computes the actuation torques for the robot’s joints. The framework is validated in simulations using the full dynamics of the robot, with different terrain types and under constant external disturbance.