Jianan Wang

Predicting task-relevant latent futures, not raw pixels, unlocks efficient and effective long-horizon VLA control, outperforming reactive policies by over 11 percentage points.

Wrist-mounted event cameras can fully replace external camera rigs in robot manipulation, surviving extreme exposure failures by using past and predicted event streams to retain off-camera spatial context.

Polyak's momentum can significantly expand the critical batch size, enabling more parallelism without compromising data efficiency.

SmoothRL achieves real-time policy optimization in robotics, effectively bridging the gap between asynchronous execution and online reinforcement learning.