Ze-Wei Zhou

Embodied AI agents can now continuously self-improve by co-evolving their policies, training curricula, and judges, overcoming the bottleneck of static evaluation.

SpanVLA slashes action generation latency in autonomous driving by bridging VLMs with flow-matching, while also learning from driving mistakes to boost robustness.

Autonomous driving policies that ace open-loop tests can still crash and burn in the real world, because they learn to ignore the reactive nature of closed-loop environments.