E-VLA: Event-Augmented Vision-Language-Action Model for Dark and Blurred Scenes

arXiv cs.RO / 4/7/2026

💬 OpinionSignals & Early TrendsIdeas & Deep AnalysisModels & Research

共有:

Key Points

The paper introduces E-VLA, an event-augmented vision-language-action (VLA) framework designed to improve robotic manipulation robustness in extreme low-light and motion-blurred conditions where frame-based perception fails.
Instead of reconstructing images from event camera data, E-VLA directly uses motion/structural cues from event streams to maintain semantic perception and perception-action consistency under sensing degradations.
The authors build an open-source teleoperation platform using a DAVIS346 event camera and collect a real-world synchronized RGB–event–action manipulation dataset across multiple tasks and illumination settings.
Experiments show large gains from event integration, including overlay fusion improving Pick-Place success in very low light (20 lux) and the event adapter reaching even higher robustness under both dark and severe motion blur.
The work also proposes lightweight, pretrained-compatible event integration and studies event windowing/fusion strategies aimed at stable real-world deployment, with code and dataset planned for release.

Abstract

Robotic Vision-Language-Action (VLA) models generalize well for open-ended manipulation, but their perception is fragile under sensing-stage degradations such as extreme low light, motion blur, and black clipping. We present E-VLA, an event-augmented VLA framework that improves manipulation robustness when conventional frame-based vision becomes unreliable. Instead of reconstructing images from events, E-VLA directly leverages motion and structural cues in event streams to preserve semantic perception and perception-action consistency under adverse conditions. We build an open-source teleoperation platform with a DAVIS346 event camera and collect a real-world synchronized RGB-event-action manipulation dataset across diverse tasks and illumination settings. We also propose lightweight, pretrained-compatible event integration strategies and study event windowing and fusion for stable deployment. Experiments show that even a simple parameter-free fusion, i.e., overlaying accumulated event maps onto RGB images, could substantially improve robustness in dark and blur-heavy scenes: on Pick-Place at 20 lux, success increases from 0% (image-only) to 60% with overlay fusion and to 90% with our event adapter; under severe motion blur (1000 ms exposure), Pick-Place improves from 0% to 20-25%, and Sorting from 5% to 32.5%. Overall, E-VLA provides systematic evidence that event-driven perception can be effectively integrated into VLA models, pointing toward robust embodied intelligence beyond conventional frame-based imaging. Code and dataset will be available at https://github.com/JJayzee/E-VLA.