Daily Arxiv

This is a page that curates AI-related papers published worldwide.
All content here is summarized using Google Gemini and operated on a non-profit basis.
Copyright for each paper belongs to the authors and their institutions; please make sure to credit the source when sharing.

LATTE-MV: Learning to Anticipate Table Tennis Hits from Monocular Videos

Created by
  • Haebom

Author

Daniel Etaat, Dvij Kalaria, Nima Rahmanian, Shankar Sastry

Outline

This paper emphasizes the importance of not only physical agility but also the ability to predict an opponent's intentions and secure reaction time for competitive table tennis. While previous studies have developed real-time table tennis game systems, they either lack predictive capabilities or are limited by the size and diversity of their datasets. This paper proposes (1) a scalable system for 3D reconstruction of monocular videos of table tennis matches and (2) an uncertainty-aware controller that predicts an opponent's actions. Simulation results show that the proposed policy improves the ball return rate for high-speed hits from 49.9% to 59.0% compared to existing non-predictive policies.

Takeaways, Limitations

•
Takeaways:
◦
Experimentally demonstrating the importance of predicting opponent actions in table tennis matches.
◦
A scalable implementation of a 3D reconstruction system using monocular video.
◦
Improving ball return rates through predictive controllers that take uncertainty into account.
•
Limitations:
◦
These are results from a simulation environment and require performance verification in an actual game environment.
◦
Limitations on the size and diversity of the dataset may still exist.
◦
Further research is needed on the generalization performance of predictive controllers.
👍
Made with Slashpage