Can Vision-Language Models Assess Proxemic Risk from Egocentric Robot Images?

arXiv:2608.12515v1 Announce Type: new Abstract: Assessing proxemic danger from a robot's egocentric perspective is critical for safe embodied navigation in human environments and requires both visual and contextual reasoning. We evaluate three opensource vision-language models (VLMs) (extit{InternVL}, extit{Qwen-VL}, and extit{SmolVLM}) on the classification of egocentric robot images into four danger levels, comparing three prompting strategies and two rounds of QLoRA fine-tuning against a stratified random baseline. Without fine-tuning, all models perform near the baseline, while fine-tuning yields only modest overall improvements. However, extit{Qwen-VL} with an advanced prompt achieves substantially higher recall for high-danger cases than the other models. An analysis of person localization further shows that correct danger classification does not correspond to better spatial grounding, indicating that a model may produce a useful safety label without attending to the relevant region of the scene. These results show that current VLMs remain limited in fine-grained proxemic reasoning and spatial grounding, although targeted prompting and fine-tuning can improve high-danger detection in selected models.

Related

• Toward Low-Latency Vision-Language Models with Doubly-Correct Predictions in Egocentric Visual Understanding
• Towards Spatial Trace with Reasoning in Vision-Language Models for Robotics
• Embodied3DBench: Benchmarking Low-Level Embodied Spatial Intelligence of Vision Language Models
• Urban Risk-Aware Navigation via VQA-Based Event Maps for People with Low Vision

Source: arXiv cs.CV | 2026-08-14