AtlasVLA: Persistent World-Ego State Modeling for Vision-Language-Action Models

2026-08-06

AI Summary

This work addresses the limitations of existing vision–language–action models in partially observable, long-horizon tasks, where performance is hindered by perceptual forgetting and loss of task progress awareness. To overcome these challenges, the authors propose a persistent world–ego state model featuring a dual-memory architecture: a 4D voxel-hash-based world state memory and an ego-centric working memory. This design enables a shift from passive reactivity to active reasoning. By integrating a diffusion Transformer with the dual-memory mechanism, the method transcends the perceptual constraints of monocular wrist-mounted cameras. It achieves state-of-the-art performance on LIBERO, RLBench, and real-world environments—boosting success rates by 9.4% on LIBERO-Long and by 17.5% on real-world long-horizon tasks using only wrist-camera input.

Abstract

While Vision-Language-Action (VLA) models have advanced embodied AI, their fundamentally reactive paradigm severely limits performance in partially observable and long-horizon tasks. When restricted to a single wrist-mounted camera, they inevitably suffer from perception forgetting as objects exit the field of view, and temporal task-progress forgetting} during multi-step execution. To overcome these bottlenecks, we propose AtlasVLA, a novel framework that transitions from direct reactive manipulation to proactive reasoning through a persistent world-ego state. AtlasVLA features a dual-memory architecture: a 4D Persistent World State Memory that lifts transient 2D observations into a globally updated, voxel-hashed spatial state to resolve visual blind spots, and an Ego-Working State Memory that tracks historical ego state and task progress. By conditioning a diffusion transformer (DiT) on this joint World-Ego state, AtlasVLA enables robust spatial reasoning. Extensive evaluations across LIBERO, RLBench, and real-world benchmarks demonstrate that AtlasVLA achieves state-of-the-art performance using solely a wrist camera. Remarkably, it decisively outperforms multi-view baselines, yielding absolute success rate improvements of 9.4% on LIBERO-Long and 17.5% in real-world long-horizon tasks.

Problem

• Vision-Language-Action models
• partial observability
• long-horizon tasks
• perception forgetting
• task-progress forgetting

Innovation

• persistent world-ego state
• dual-memory architecture
• voxel-hashed spatial memory
• diffusion transformer
• long-horizon embodied reasoning