Thanh-Tuan Tran

Achieving 86.79% action classification accuracy, this framework outperforms existing methods by over 80% in generating precise robotic commands from video demonstrations.

Hybrid TD3 stabilizes reinforcement learning in hybrid action spaces by taming overestimation bias with a theoretically grounded weighted clipped Q-learning target.

Robots can now learn manipulation skills from unstructured videos with significantly improved accuracy and generalization by decoupling video understanding from policy learning.