Minimum Time Footstep Planning for Simulated Robot Soccer Kicks Using Model Predictive Control

This work contributes with a footstep planner that applies Model Predictive Control techniques, along with non-linear and mixed-integer numerical optimization, to generate footstep plans for the NAO humanoid robot to kick a soccer ball, within the context of the Robocup Soccer Simulation 3D league. The resulting planner can easily be adapted to other Robocup leagues and applications that require footstep planning. The planner accounts for several constraints related to the employed physical platform, as well as obstacles that may be present in the environment, and minimizes the amount of footsteps present. Two different approaches to find the plans with the least amount of footsteps were proposed, namely the Increasing Horizon Method (IHM) and Constraint Relaxation Method (CRM), and the resulting plans and execution times for different parameters were studied. The footstep plans generated using both methods display interesting emergent behaviors, such as rotating a foot to extend the next footstep's reach, or placing footsteps in a way that, at the end of the plan, the robot faces a completely different direction than at the start.