STARS: From Spatiotemporal Dynamics to Social Representations in Human-Robot Interaction
SocialNav-SUB benchmarks whether vision-language models understand social robot navigation scenes.
STARS introduces SocialNav-SUB, a visual question-answering benchmark for scene understanding in social robot navigation. It tests spatial, spatiotemporal, and social reasoning against human and rule-based baselines. Experiments find the best vision-language model still trails simpler rule-based methods and human consensus, exposing gaps in current models' social scene understanding. Code and data are released with the paper.
- VQA tasks cover spatial, temporal, and social reasoning.
- Best VLM still trails rule-based and human baselines.
- Benchmark, code, and data are publicly released.
From the original article
1 min read at arxiv.orgRobot navigation in dynamic, human-centered environments requires socially-compliant decisions grounded in robust scene understanding. Recent Vision-Language Models (VLMs) exhibit promising capabilities such as object recognition, common-sense reasoning, and contextual understanding, capabilities that align with the nuanced requirements of social robot navigation.
By Nathan Tsoi · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.