News

Lute presented OpenGVL at the NeurIPS 2025 EWM Workshop

We attended the NeurIPS 2025 Workshop on Embodied World Models for Decision Making with OpenGVL, our benchmark for visual temporal progress tracking in robotics data.

OpenGVL studies how well vision-language models estimate task progress across robotic and human demonstrations. The benchmark is designed to support data curation by identifying whether a trajectory is moving toward task completion.

The work highlights the gap between closed-source and open-source models on progress tracking, and shows how temporal progress signals can help filter and evaluate large robotics datasets.