Model predictive control approach to enhance stable walking for planar bipedal robots

Details

Supervisors

Jungers, Raphaël M.

Faculty

Ecole polytechnique de Louvain

Degree label

Master [120] : ingénieur civil électromécanicien, à finalité spécialisée: mécatronique

Abstract

enMotivated by the intrinsic control challenges in bipedal robotics, this thesis explores two Model Predictive Control (MPC) approaches. Both rely on the Zero Moment Point (ZMP) criterion, pre-planned footholds, and state feedback to generate stable walking patterns for fully actuated planar robots. The first approach employs the Linear Inverted Pendulum Model (LIPM) as a simplified representation to meet the typical time constraints imposed by embedded hardware. The second approach incorporates the robot’s full nonlinear dynamics to optimise actuator commands, thereby enhancing walking efficiency while increasing computational complexity. Both approaches are implemented in Julia and evaluated through simulations. Each component of the control architectures, including the footstep and trajectory planner, discrete-time MPC formulation, inverse kinematics solver, and low-level controller, is described in detail. Walking stability, actuator feasibility, and computational performance are thoroughly discussed. Overall, this thesis provides a solid foundation for future MPC-based research on underactuated bipedal walking robots and a closed-loop benchmark for the UCLouvain Learning to Control project software, Dionysos.jl.