Host: The Japan Society of Mechanical Engineers
Name : [in Japanese]
Date : November 15, 2024 - November 17, 2024
In an outside pass in soccer, we observe four strategies: the backswing, the stopping of the leg, the guiding of the foot outward as the knee joint extends, and the stopping of the follow-through. The aim of this research is to validate if we observe these four strategies by training the agent with a reward focused on kicking a faster ball. Additionally, we aim to investigate the contribution of the collateral ligaments in the outside pass. We built two leg models: (a) a two-degree-of-freedom leg model that can apply roll torque around the hip joint, and (b) a three-degree-of-freedom leg model that can apply pitch-roll-yaw torque around the hip joint and pitch torque around the knee joint. We trained a Deep Deterministic Policy Gradient (DDPG) agent using these two leg models and analyzed the torque around the hip and knee joints, as well as the ball velocity after the leg loses contact with the ball. In the (a) two-degree-of-freedom leg model, the agent shows the stopping the follow-through strategy at the episode 1826, and it shows both the backswing strategy and stopping the follow-through strategies at episode 3000. In the (b) three-degree-of-freedom leg model, the agent shows the guiding of the foot outward as the knee joint extends and the stopping follow-through at episode 2127, and the same strategies were observed at episode 3000. Leg model (a), which utilizes the collateral ligaments’ stiffness resulted in a faster ball compared to leg model (b), which is capable of moving in directions that do not engage the collateral ligaments’ stiffness. This suggests that utilizing collateral ligaments’ stiffness has advantage for kicking faster balls in an outside pass. This findings will lead to finding new collateral ligaments’ functions.