SOURCE-LINKED INTELLIGENCE

SimpleMemVLA: A Simple but Effective Native-Video Memory for Vision-Language-Action Models

arXiv · AI, language, vision and robotics · article · Sep 2, 2026 · UTC

Long-horizon manipulation is partially observable: the information needed to choose the next action may appear only in observations from minutes earlier. Existing memory mechanisms: retrieval banks, learned compressors, recurrent states must decide what to keep from the past before knowing what a future decision will require. This was motivated by the assumption that minute-scale history is too large to process directly, which modern VLM backbones no longer make true. In this work, we introduce SimpleMemVLA, a VLA without a dedicated memory module. It keeps the sampled history intact and passe

Read original source ↗ Open in workspace

recordType

paper

region

Global