Yutong Hu

2AM shows that task memory can remain Agent-side and shows that Action Model capability depends not only on what the policy has learned, but on how precisely the Agent can steer it, and isolates this question through a deliberately constrained design.

Object-object interactions can dramatically reduce the training data needed for effective manipulation policies, achieving superior performance with less complexity.

Assistron achieves a remarkable balance between autonomy and user control, significantly enhancing task success while reducing user effort in daily activities.

FF-JEPA transforms long-horizon planning by enabling goal-free trajectory optimization without the need for explicit goal images.

Achieve robust long-horizon visual control by adaptively balancing model-based lookahead with bootstrapping, enabling zero-shot transfer to real-world tasks with severe occlusions.

A million-scale dataset of globally diverse, cross-modal geo-location pairs, coupled with a novel physical-law-aware network, leapfrogs existing CMGL benchmarks and opens the door to truly universal positioning systems.