Research
We partner with the ICON Lab at Columbia University to research on hazardous robotics
Zhengbo Zou
Chief Scientist, Cervo · Assistant Professor of Civil Engineering and Engineering Mechanics, Columbia University · Director, ICON Lab
Research directions
The first hazardous benchmark
Knowledge-check and action-check: can a model understand a hazardous task, and can it perform it under site conditions.
VLA labelling evaluation
How well current vision-language-action models identify task, environment, tools and task boundaries in raw hazardous footage.
Annotation pipelines for hazardous environments
Labelling schemas and expert-in-the-loop workflows that turn a shift of footage into training-ready episodes.
Open-source models for hazardous tasks
Robotic models fine-tuned on hazardous-work data and released for the industry.