Researchers have developed a new method called LATENT that enables humanoid robots to learn complex tennis skills from imperfect human motion capture data. The system combines imitation learning with reinforcement learning to achieve robust athletic performance including forehands, backhands, and overhead smashes. This represents significant progress in transferring dynamic human movements to humanoid robots.
Background
Humanoid robotics has long sought to replicate complex human athletic movements, but transferring dynamic sports skills from human demonstrations to robots remains challenging due to hardware limitations and imperfect training data.
- Source
- Hacker News (RSS)
- Published
- Mar 15, 2026 at 11:21 PM
- Score
- 7.0 / 10