metadata
license: mit
task_categories:
- video-text-to-text
tags:
- robotics
- human-robot-interaction
- safety
TouchSafeBench
This repository contains the data presented in Probing Collision Grounding in Vision-Language Models for Safe Human-Robot Collaboration.
TouchSafeBench is a physics-grounded benchmark for evaluating collision grounding in vision-language models (VLMs). Built in Habitat 3.0, it contains 2,940 simulated indoor co-presence episodes across social navigation and social rearrangement, with synchronized multi-view RGB-D observations, top-down trajectory maps, calibrated camera metadata, and simulator-derived contact labels.
- Project page: https://jun25336-bit.github.io/TouchSafeBench/
- Code: https://github.com/jun25336-bit/TouchSafeBench
The benchmark supports two deployment-facing tasks:
- Safety event classification: no event, scene collision, or human collision risk.
- Collision warning: no risk, operational hazard, or catastrophic risk.