EPIC-Bench: A Perception-Centric Benchmark for Fine-Grained Embodied Visual Grounding in Vision-Language Models

2026-05-16

AI Summary

Existing vision-language model benchmarks predominantly rely on question-answering formats, which are susceptible to linguistic priors and inadequately assess fine-grained visual grounding capabilities in embodied scenarios. To address this limitation, this work introduces EPIC-Bench—the first perception-centric benchmark specifically designed to evaluate fine-grained visual alignment in real-world physical interactions. It comprises 6.6K image-text-mask triplets spanning 23 tasks across three stages: object localization, navigation, and manipulation. Emphasizing mask-level annotations and task diversity, EPIC-Bench enables systematic evaluation of 89 state-of-the-art models, revealing critical bottlenecks in multi-object counting, part-whole reasoning, and identification of actionable regions. This benchmark thus provides a robust foundation and clear direction for advancing perceptual modules in embodied intelligence.

Abstract

While large vision-language models (VLMs) are increasingly adopted as the perceptual backbone for embodied agents, existing benchmarks often rely on question-answering or multiple-choice formats. These protocols allow models to exploit linguistic priors rather than demonstrating genuine visual grounding. To address this, we present EPIC-Bench, Embodied PerceptIon BenChmark, a fine-grained grounding benchmark designed to systematically evaluate the visual perceptual capabilities of VLMs in real-world embodied environments. Comprising 6.6k meticulously annotated tuples (Image, Text, Mask), EPIC-Bench spans 23 fine-grained tasks across three core stages of the embodied interaction pipeline: Target Localization, Navigation, and Manipulation. Extensive evaluations of over 89 leading VLMs reveal that while advanced reasoning models show promise, current VLMs universally struggle with complex visual-text alignment for physical interactions. Specifically, models exhibit critical bottlenecks in multi-target counting, part-whole relationship understanding, and affordance region detection. EPIC-Bench provides a robust foundation and actionable insights for advancing the next generation of vision-driven embodied models.

Problem

• visual grounding
• vision-language models
• embodied AI
• perception benchmark
• fine-grained evaluation

Innovation

• embodied visual grounding
• perception-centric benchmark
• fine-grained evaluation
• vision-language models
• visual-text alignment