Inverse Kinematics Kicking in the Humanoid RoboCup Simulation League

This paper describes the inverse kinematics kicking system used in UT Austin Villa’s submission to the humanoid RoboCup simulation league competition. It describes the infrastructure we designed and how machine learning is used to optimize a set of kicks with the hope of developing a versatile and omnidirectional kick. As of now, the ability to kick has not substantially affected the outcome of the 3D simulation competition as walk speed and dribbling have been the main focus and the strongest factors in a team’s ability to do well competitively. The underlying goal is to create a set of kicks that can be optimized using machine learning to be faster to align but that can achieve similar distance and accuracy to fix keyframe based skills. To evaluate this goal we look to test the kick over many iterations against varied opponents and situations. We hypothesize that it is possible to create an abstracted representation of skills such as kicking through the use of inverse kinematics based curves that while simultaneously achieving the performance goals we pose and allowing a much simpler human interaction with the system.