Search papers, labs, and topics across Lattice.
This paper introduces AtlasVLA, a novel framework designed to enhance Vision-Language-Action (VLA) models by transitioning from reactive to proactive reasoning through a persistent world-ego state. By employing a dual-memory architecture that includes a 4D Persistent World State Memory and an Ego-Working State Memory, AtlasVLA effectively mitigates perception and task-progress forgetting, enabling robust spatial reasoning even with a single wrist-mounted camera. Experimental results demonstrate that AtlasVLA achieves state-of-the-art performance, surpassing multi-view baselines with significant improvements in success rates on both simulated and real-world long-horizon tasks.
AtlasVLA outperforms multi-view baselines by over 17% in long-horizon tasks, showcasing the power of proactive reasoning in embodied AI.
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.
