An Improved Formulation for Model Predictive Control of Legged Robots for Gait Planning and Feedback Control

Abstract

Predictive control methods for walking commonly use low dimensional models, such as a Linear Inverted Pendulum Model (LIPM), for simplifying the complex dynamics of legged robots. This paper identifies the physical limitations of the modeling methods that do not account for external disturbances, and then analyzes the issues of numerical stability of Model Predictive Control (MPC) using different models with variable receding horizons. We propose a new modeling formulation that can be used for both gait planning and feedback control in an MPC scheme. The advantages are the improved numerical stability for long prediction horizons and the robustness against various disturbances. Benchmarks were rigorously studied to compare the proposed MPC scheme with the existing ones in terms of numerical stability and disturbance rejection. The effectiveness of the controller is demonstrated in both MATLAB and Gazebo simulations.

Bayesian Optimization for Whole-Body Control of High-Degree-of-Freedom Robots Through Reduction of Dimensionality

“…An episode consists of a start and an end phase of 1s each and 5 straight steps of 0.25m with a step duration of 3s. The walking motion is generated by using Model-Predictive Control as in [34] to track these pre-planned footsteps for both Gait Planning and Feedback Control. At a frequency of 50Hz, the sum of 850 rewards (4) are gathered per episode as the scalar output of the reward function.…”

Model Predictive Control for Motion Planning of Quadrupedal Locomotion

“…So in this study, we are going to apply the method of bipedal gait generation for quadrupedal planning. In bipedal locomotion, a promising approach to generate walking mo- tions online is to use model predictive control (MPC) method for autonomous walking [15], [16], [17], [18], [19], [20]. Considering the task commands and the system constraints, MPC-based scheme can generate optimal trajectories within the predictive horizon, according to the current state of the system [21].…”

Learning Whole-Body Motor Skills for Humanoids

“…First, supplementing existing control strategies with learning methods allows dealing with scenarios that are hard to engineer in a traditional sense such as sudden, high impact forces and discrete, sudden switches of contact. Second, in contrast to planning and control algorithms [1]- [3] that demand high computational power to run at or close to real-time, e.g. Model-Predictive Control, the computation for machine learning approaches can be outsourced offline.…”

“…Email: chuanyu.yang@ed.ac.uk. Second, in contrast to planning and control algorithms [1]- [3] that demand high computational power to run at or close to real-time, e.g. Model-Predictive Control, the computation for machine learning approaches can be outsourced offline.…”