Researchers have developed a new framework called AdaPT to enable humanoid robots to perform professional tennis styles, focusing on adaptive motion planning and tracking. This system learns from broadcast videos to generate stylistic kinematic motions, which are then executed by a tracker. To bridge the gap between simulation and real-world performance, AdaPT includes an adaptation mechanism that improves tracking robustness by learning to handle randomized execution speeds and conditions the planner on a learned motion-speed adapter to reduce compounding errors. Experiments on the Unitree G1 and the larger Dobot Atom humanoid robots have demonstrated the effectiveness of this approach in achieving in-the-wild serving capabilities without motion capture. AI
IMPACT Enables humanoid robots to perform complex, professional-level athletic tasks, advancing robotics capabilities in dynamic environments.
RANK_REASON Academic paper detailing a new robotics framework and its experimental validation. [lever_c_demoted from research: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
