Researchers from Tsinghua University, Peking University, and collaborating institutions released LATENT (Learning Athletic humanoid TEnnis skills from imperfect human motioN daTa) on March 13, 2026, a framework that teaches humanoid robots to play tennis using noisy, amateur motion clips instead of expensive professional motion capture data. The system addresses a fundamental bottleneck in robotic learning: the need for perfectly labeled, high-quality movement data from professional athletes.
Framework Processes Scattered, Low-Quality Human Movement Clips
LATENT's core innovation lies in its ability to learn from imperfect data sources including amateur swings, partial sequences, and low-resolution video. The framework provides an open-source pipeline covering motion tracker pre-training, online distillation, and high-level policy learning. Playing tennis requires robots to track the ball early, position themselves correctly, select appropriate shots, and maintain balance while swinging—demanding dynamic movements, agile whole-body coordination, and rapid reactions.
The technical implementation includes:
- Motion tracker pre-training for interpreting noisy input data
- Online distillation to refine learned behaviors
- High-level policy learning for strategic decision-making
- Real-time adaptation rather than replaying pre-programmed sequences
Open-Source Release Includes Code and Training Data
The research team released the tracking codebase and a subset of human tennis motion data on GitHub on March 13, 2026. The project page at zzk273.github.io/LATENT provides additional documentation and demonstrations. Related demonstrations include UBTech's Walker S2 humanoid robot wielding a tennis racket with precision and moving dynamically around the court, suggesting effective whole-body control systems actively adapting in real time.
The research generated significant community interest on Hacker News, receiving 119 points and 24 comments. Lead researcher Zhikai Zhang described the work as "a step toward athletic humanoid sports skills" that enables dynamic movements and rapid reactions through learning from imperfect human motion data.
Key Takeaways
- LATENT framework released March 13, 2026 by Tsinghua University, Peking University, and collaborators
- System learns tennis skills from noisy, amateur motion clips instead of expensive professional motion capture data
- Framework includes motion tracker pre-training, online distillation, and high-level policy learning
- Open-source code and training data available on GitHub under MIT license
- Research received 119 points and 24 comments on Hacker News within days of release