Latest writing
Si0: The Capabilities That Transfer, Perceived Through the Robot's Eyes
The capabilities, data, and sensor alignment behind Si0 as a spatial-intelligence backbone for humanoids.
ROVE: Unlocking Human Interventions for Humanoid Manipulation via Reinforcement Learning
The first human-in-the-loop VLA-RL framework for humanoid robots with dexterous hands.
AGRA: Making Foresight Actionable in World Action Models
An action-grounded representation alignment objective that bridges visual prediction and robot control in World Action Models.
Inside Fe0: What Cross-Embodiment Data Teaches an Embodied Foundation Model
A close look at how heterogeneous robot and human data helps an embodied foundation model generalize across visual, semantic, relational, planning, and control axes.
UniT: Toward a Unified Physical Language for Human-to-Humanoid Policy Learning and World Modeling
A unified physical language that maps heterogeneous human and robot motion into shared action primitives.
DIAL: Decoupling Intent and Action via Latent World Modeling for End-to-End VLA
Latent world modeling that decouples high-level intent from low-level actions in end-to-end vision-language-action policies.