Yi Yu

ARENA uncovers vulnerabilities in large audio-language models that traditional text-based red-teaming methods miss, achieving near-perfect safety metrics across multiple systems.

Organizing demonstrations into a simple-to-complex structure can dramatically boost the efficiency and stability of robotic manipulation learning.

Achieving consistent voxel spacing can dramatically improve medical image segmentation quality, leading to better diagnostic and surgical outcomes.

Transitioning from predictive to epistemic intelligence could revolutionize how embodied systems learn and adapt to complex environments.

Execution instability is the leading failure mode in low-cost robotic manipulation, revealing critical gaps in VLA model robustness that traditional success metrics overlook.