Lei Zhu

A single adversarial texture can compromise the performance of Vision-Language-Action models across multiple tasks, revealing alarming shared vulnerabilities.

Rarity-aware sampling and spatial consistency techniques enable discrete diffusion to achieve high-quality image super-resolution with fewer decoding steps than traditional methods.

Quantizing VLAs for robots doesn't have to trash performance: DA-PTQ recovers near full-precision accuracy even at low bit-widths by explicitly minimizing kinematic drift during sequential control.

Forget relying on immediate observations: Keyframe-Chaining VLA lets robots nail long-horizon tasks by remembering and chaining together only the *important* past states.