MAKP: Multi-Mode Accurate Kicking Policy for Humanoid Robots
Zheng Zhang, Kaiyang Xu, Zhanxiang Cao, Yizhi Chen, Peng Wang, Haoyang Li, Yang Zhang, Shengcheng Fu, Xin Shen, Xiaokang Yang, Yue Gao
AI summary
Problem
Humanoid robots struggle to balance stable single-leg support with accurate ball trajectory control during kicking, as traditional methods lack target-specific precision and existing learning approaches fail to reconcile motion naturalness with kicking performance.
Approach
MAKP combines a Motion Diffusion Model to generate diverse kicking trajectories with a three-stage reinforcement learning strategy that progressively trains motion tracking, kicking accuracy, and adaptive balance via a multi-critic architecture.
Key results
- Generates diverse, human-like kicking motions via diffusion models
- Achieves high-precision ball targeting across varied angles and distances
- Maintains stable single-leg balance during dynamic kicking phases
- Validates successful sim-to-sim and sim-to-real transfer on the Booster T1 platform
Why it matters
Advances the deployment of bipedal robots in dynamic sports and manipulation by providing a robust, learning-based framework that balances motion fidelity with task accuracy.
Abstract
Humanoid robot soccer players face fundamen- tal challenges in achieving stable motion execution and ball trajectory control, particularly under balance constraints dur- ing single-leg support phases. In this paper, we introduce MAKP (Multi-mode Accurate Kicking Policy), a novel motion generation-based end-to-end kicking paradigm that enables humanoid robots to perform accurate ball kicking while ex- ecuting diverse kicking motions. MAKP uniquely integrates a diffusion-based motion generator to produce various kick- ing trajectories and employs a three-stage learning strategy to address the inherent trade-off between motion similarity and kicking performance. Stage I focuses on stable motion tracking and single-leg balance maintenance, while Stage II optimizes ball kicking capabilities. In Stage III, we introduce a Multi-Critic mechanism combined with curriculum learning to further enhance the balance between kicking accuracy, motion similarity and robot stability. Real-world experiments on the Booster T1 platform validate the effectiveness of our approach.