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

TL;DR

EPIC-Bench is a comprehensive, fine-grained benchmark designed to evaluate the visual grounding capabilities of vision-language models in real-world embodied tasks, highlighting current limitations.

Contribution

The paper introduces EPIC-Bench, a new benchmark with 6.6k annotated tuples across 23 tasks, specifically targeting visual perception in embodied environments.

Findings

Current VLMs struggle with complex visual-text alignment for physical interactions.

Models show bottlenecks in multi-target counting and part-whole understanding.

Advanced reasoning models still face significant challenges in embodied visual grounding.

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.