EventVLA: Event-Driven Visual Evidence Memory for Long-Horizon Vision-Language-Action Policies

2026-06-18

AI Summary

This work addresses the challenge of long-horizon robotic manipulation, where task-critical visual cues are often occluded or become unobservable, leading to failure in existing vision-language-action (VLA) policies. To overcome this limitation, the authors propose EventVLA, an end-to-end framework that introduces a novel foresight mechanism: it predicts the probability of future keyframes directly from VLA latent representations, enabling dynamic assessment of the causal utility of current observations. By integrating visual anchors with a Keyframe Evidence Memory (KEM) module, EventVLA proactively preserves sparse yet crucial visual evidence before it disappears. Evaluated on the newly introduced non-Markovian RoboTwin-MeM benchmark, the method achieves a 40% average improvement in success rate over the current best memory-augmented VLA approaches across 17 simulated and 4 real-world dual-arm tasks.

Abstract

Memory remains a critical bottleneck for long-horizon robotic manipulation, as standard Vision-Language-Action (VLA) policies often fail when task-relevant cues become occluded or unobservable over time. While existing memory-augmented methods utilize historical context, they either suffer from severe information bottlenecks, incur high latency via decoupled dual systems, or rely on unselective buffers that accumulate massive visual redundancies. To address these limitations, we introduce EventVLA, an end-to-end framework founded on the concept of sparse visual evidence memory that comprises two core components: foundational visual anchors to retain initial and short-term contexts, and a dynamic Keyframe Evidence Memory (KEM) module. Specifically, KEM directly predicts future keyframe probabilities from the VLA's latent embeddings to autonomously capture and store sparse, task-critical visual events. This foresight-driven mechanism empowers the policy to dynamically evaluate the future causal utility of current observations, preserving transient visual evidence before it becomes unobservable. Furthermore, we propose RoboTwin-MeM, a diagnostic benchmark specifically designed to evaluate non-Markovian manipulation tasks with interactive visual evidence. Extensive evaluations show that across 17 memory-requiring simulation tasks and 4 real-world bimanual tasks, EventVLA achieves an average success rate improvement of +40% over state-of-the-art memory-augmented VLAs.

Problem

• long-horizon manipulation
• visual evidence memory
• occlusion
• non-Markovian tasks
• memory bottleneck

Innovation

• Event-Driven Memory
• Keyframe Evidence Memory
• Vision-Language-Action
• Sparse Visual Evidence
• Long-Horizon Manipulation