Xinyu Cui

A maritime navigation policy trained purely in simulation can now successfully pilot a 17-ton vessel in real-world congested waters, zero-shot.

Quadrupedal robots can now swim more efficiently and stably thanks to a novel constrained reinforcement learning approach that tames destabilizing forces in complex fluid environments.

Flow-based VLAs can now learn online without likelihoods or value networks, unlocking better generalization in complex embodied control tasks.