ReMem-VLA: Empowering Vision-Language-Action Model with Memory via Dual-Level Recurrent Queries

TL;DR

ReMem-VLA introduces a dual-level recurrent memory system with learnable queries for vision-language-action models, significantly improving long-term and short-term memory retention in robot control tasks.

Contribution

The paper proposes ReMem-VLA, a novel memory-augmented VLA model with learnable recurrent queries for enhanced temporal memory without extra inference costs.

Findings

ReMem-VLA outperforms memory-free baselines on memory-dependent tasks.

It demonstrates strong spatial, sequential, episodic, temporal, and visual memory capabilities.

The model achieves significant improvements in real-world robot experiments.

Abstract

Vision-language-action (VLA) models for closed-loop robot control are typically cast under the Markov assumption, making them prone to errors on tasks requiring historical context. To incorporate memory, existing VLAs either retrieve from a memory bank, which can be misled by distractors, or extend the frame window, whose fixed horizon still limits long-term retention. In this paper, we introduce ReMem-VLA, a Recurrent Memory VLA model equipped with two sets of learnable queries: frame-level recurrent memory queries for propagating information across consecutive frames to support short-term memory, and chunk-level recurrent memory queries for carrying context across temporal chunks for long-term memory. These queries are trained end-to-end to aggregate and maintain relevant context over time, implicitly guiding the model's decisions without additional training or inference cost. Furthermore, to enhance visual memory, we introduce Past Observation Prediction as an auxiliary training objective. Through extensive memory-centric simulation and real-world robot experiments, we demonstrate that ReMem-VLA exhibits strong memory capabilities across multiple dimensions, including spatial, sequential, episodic, temporal, and visual memory. ReMem-VLA significantly outperforms memory-free VLA baselines $\pi$0.5 and OpenVLA-OFT and surpasses MemoryVLA on memory-dependent tasks by a large margin.