Jia Shi

Internet videos unlock scalable, real-world navigation data, enabling robots to achieve 94.9% success on complex language-conditioned tasks.

MachEmbodied-U0, a unified embodied foundation model connecting understanding and generation experts through a Mixture-of-Transformers architecture, is presented, a unified embodied foundation model connecting understanding and generation experts through a Mixture-of-Transformers architecture that combines competitive downstream control performance with transferable task-grounding and visual-dynamic capabilities across simulation and the real world.

Class-name descriptors can mislead vision-language models, but selecting attributes directly from images boosts accuracy and interpretability significantly.