Yichen Zhu

UMMs exhibit significant exposure bias in multi-turn interactions, revealing critical performance gaps that existing benchmarks overlook.

ActiveMimic reveals that leveraging active perception from egocentric videos can close the performance gap with robot-pretrained models, transforming how we approach robot learning.

Forget slow, expensive real-world trials: dWorldEval's discrete diffusion world model lets you evaluate robot policies across thousands of environments and tasks with unprecedented speed and accuracy.

Forget expensive real-world robot training: Hi-WM lets humans directly edit a robot's simulated reality, turning world models into powerful, reusable playgrounds for failure recovery.