Tongyu Wu

1,000 hours of human touch data unlocks a 37% jump in vision-only dexterous manipulation success, revealing touch as a powerful pretraining signal beyond its direct use in policies.

Adversarial training with deep reinforcement learning cracks the computationally intractable Maximal Covering Location-Interdiction Problem, outperforming traditional methods in speed and solution quality.

DRL agents can now solve vehicle routing problems across diverse real-world distributions thanks to a new architecture that adaptively recombines specialized policy modules.