Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

A Biomechanics-Inspired Approach to Soccer Kicking for Humanoid Robots

Daniel Marew, Nisal Perera, Shangqun Yu, Sarah Roelker, Donghyun Kim

TL;DR

This paper tackles the challenge of dynamic, powerful soccer kicking for humanoid robots by introducing a biomechanics-informed framework that fuses kinodynamic trajectory optimization with imitation learning. It leverages motion-capture data retargeted to a 25 DoF humanoid (PresToe) to generate physically feasible, momentum-building kick trajectories, then trains a PPO policy to robustly track these references under full-body dynamics. The approach yields instep kicks exceeding $11~m/s$ in simulation, approximately doubling the performance of prior ground-anchored methods and reaching around 40% of human capability, while respecting torque and contact constraints. A key finding is that kinodynamic planning provides a better, sample-efficient foundation for RL than purely kinematic references, and a reference-guided early termination strategy further enhances training efficiency.

Abstract

Soccer kicking is a complex whole-body motion that requires intricate coordination of various motor actions. To accomplish such dynamic motion in a humanoid robot, the robot needs to simultaneously: 1) transfer high kinetic energy to the kicking leg, 2) maintain balance and stability of the entire body, and 3) manage the impact disturbance from the ball during the kicking moment. Prior studies on robotic soccer kicking often prioritized stability, leading to overly conservative quasi-static motions. In this work, we present a biomechanics-inspired control framework that leverages trajectory optimization and imitation learning to facilitate highly dynamic soccer kicks in humanoid robots. We conducted an in-depth analysis of human soccer kick biomechanics to identify key motion constraints. Based on this understanding, we designed kinodynamically feasible trajectories that are then used as a reference in imitation learning to develop a robust feedback control policy. We demonstrate the effectiveness of our approach through a simulation of an anthropomorphic 25 DoF bipedal humanoid robot, named PresToe, which is equipped with 7 DoF legs, including a unique actuated toe. Using our framework, PresToe can execute dynamic instep kicks, propelling the ball at speeds exceeding 11m/s in full dynamics simulation.

A Biomechanics-Inspired Approach to Soccer Kicking for Humanoid Robots

TL;DR

This paper tackles the challenge of dynamic, powerful soccer kicking for humanoid robots by introducing a biomechanics-informed framework that fuses kinodynamic trajectory optimization with imitation learning. It leverages motion-capture data retargeted to a 25 DoF humanoid (PresToe) to generate physically feasible, momentum-building kick trajectories, then trains a PPO policy to robustly track these references under full-body dynamics. The approach yields instep kicks exceeding in simulation, approximately doubling the performance of prior ground-anchored methods and reaching around 40% of human capability, while respecting torque and contact constraints. A key finding is that kinodynamic planning provides a better, sample-efficient foundation for RL than purely kinematic references, and a reference-guided early termination strategy further enhances training efficiency.

Abstract

Soccer kicking is a complex whole-body motion that requires intricate coordination of various motor actions. To accomplish such dynamic motion in a humanoid robot, the robot needs to simultaneously: 1) transfer high kinetic energy to the kicking leg, 2) maintain balance and stability of the entire body, and 3) manage the impact disturbance from the ball during the kicking moment. Prior studies on robotic soccer kicking often prioritized stability, leading to overly conservative quasi-static motions. In this work, we present a biomechanics-inspired control framework that leverages trajectory optimization and imitation learning to facilitate highly dynamic soccer kicks in humanoid robots. We conducted an in-depth analysis of human soccer kick biomechanics to identify key motion constraints. Based on this understanding, we designed kinodynamically feasible trajectories that are then used as a reference in imitation learning to develop a robust feedback control policy. We demonstrate the effectiveness of our approach through a simulation of an anthropomorphic 25 DoF bipedal humanoid robot, named PresToe, which is equipped with 7 DoF legs, including a unique actuated toe. Using our framework, PresToe can execute dynamic instep kicks, propelling the ball at speeds exceeding 11m/s in full dynamics simulation.
Paper Structure (14 sections, 27 equations, 7 figures, 2 tables)

This paper contains 14 sections, 27 equations, 7 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Methodology
  3. Offline Kicking Trajectory Generation
  4. Mocap Data Preprocessing
  5. Kinodynamic Trajectory Optimization
  6. Imitation Learning
  7. Sample Efficient RL Training Through Reference Guided Early Termination Strategy
  8. Experimental Setup
  9. Experimental Platform
  10. Reference Motion Generation and RL Policy Training
  11. Sample Efficiency Experiment
  12. Ball Kicking Experiment
  13. Results and Discussion
  14. Conclusion and Future work

Figures (7)

  • Figure 1: Phases of an in-step soccer kick: (a) Approach/Run-up, (b) Planting/Support, (c) Wind-up/Backswing, (d) Cocking, (e) Swing/Acceleration, and (f) Follow-through. Adapted from bio_youtube_video.
  • Figure 2: PresToe: (a) Robot degrees of freedom (b) Foot contact points
  • Figure 3: Offline Motion Planning Steps: (a) MoCap Link Rescaling: Corresponding joints are highlighted in the same color (on the MoCap skeleton in (a) and robot in (b)). For clarity, only the left-side joint correspondence between the MoCap data and the robot is shown. (b) Inverse kinematics-based retargeting, ignoring dynamics. (c) Kinodynamic trajectory optimization. (d) Full-body dynamics-based torque limit violation verification.
  • Figure 4: Imitation learning framework
  • Figure 5: Stair Case Distance Threshold Function
  • ...and 2 more figures