Yunke Wang

Losing a single camera feed derails state-of-the-art VLA models, but pairing action-conditioned world model extrapolation with selective view withdrawal keeps closed-loop manipulation intact during sensor failures.

Object-conditioned cues can dramatically enhance 3D reconstruction quality, especially under tight frame budgets, outperforming traditional geometry-focused strategies.

StellaVLA achieves a remarkable 98.8% success rate in adapting VLA models to out-of-distribution tasks using structured demonstrations without human annotation.

Achieving a 3.2x speedup in video diffusion without sacrificing fidelity could redefine efficiency benchmarks in the field.