Zhipeng Zhang

A lightweight latent world model directly conditions VLA action generation on predicted future consequences, boosting real-world robotic success rates from 38.3% to 73.3% without pixel-level video prediction.

A novel degenerate dissipation mechanism allows viscosity in one phase to drive the dynamics of an inviscid phase in the low Mach number limit of two-phase flow.

GSR enables lightweight models to outperform heavily scaled counterparts, achieving near-perfect robustness to instruction rewording.