MemoryVLA++: Temporal Modeling via Memory and Imagination in Vision-Language-Action Models

arXiv 2026 58 method

TLDR

MemoryVLA++ integrates memory and a world model into VLA models for temporally consistent robotic manipulation.

Reasoning

The paper presents a novel framework combining working memory, episodic memory, and a world model for future state imagination, addressing temporal modeling in robotic tasks. Strengths include extensive experiments on simulation and real robots. Weaknesses are limited architectural details of the world model and potential reliance on pretrained VLMs.

Read-first score

Read-first score 58, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 35.

Recency 6%

100

Uses a gentle age decay so recent papers surface without erasing older foundations. 2026

Citation impact 18%

Uses OpenAlex-shaped citation metadata as a bibliometric attention signal, separate from paper quality. citation_normalized_percentile=0.95976633

Methodology quality 18%

Screens visible abstract and analysis fields for experiment, dataset, baseline, metric, and limitation evidence. markers=benchmark,experiment,result

Topical relevance 29%

Uses existing LLM keyword relevance scores normalized to 0-100. world model,world simulator,generative world model,interactive world model,video world model,world dynamics prediction,model-based reinforcement learning world model

Reproducibility 18%

Screens links and visible text for paper, code, dataset, artifact, and repository signals. pdf=True; code=False; dataset=False; markers=code,github

Citation velocity 12%

Citation velocity estimates citations per publication-year to reduce old-paper bias. velocity=0.00

Field roles

FoundationFrontierBridgeMethodology anchor

Rank sensitivity

Stability: volatile; rank range: 415.

Keyword Scores

world model

world dynamics prediction

generative world model

model-based reinforcement learning world model

world simulator

interactive world model

video world model

Deep Analysis

Innovations

Introduces a full temporal modeling framework for VLA models inspired by cognitive science, incorporating working memory, episodic memory via a Perceptual-Cognitive Memory Bank, and imagination through a world model for future state prediction.
Proposes redundancy-aware consolidation for updating the memory bank, and integrates imagined latents under memory guidance to form temporal-aware tokens for action prediction.

Methodology

MemoryVLA++ uses a pretrained VLM to encode current observations into perceptual and cognitive tokens (working memory). These tokens query a Perceptual-Cognitive Memory Bank that stores low-level details and high-level semantics from past interactions, updated via redundancy-aware consolidation. A world model imagines future states in a denoising latent space, and the imagined latents are integrated under memory guidance to form full temporal-aware tokens, which then condition a diffusion action expert to predict temporally consistent action sequences.

Key Results

The method achieves strong performance across 5 simulation benchmarks (Libero, SimplerEnv, Mikasa-Robo, Calvin, Libero-Plus) and 3 categories of real-robot tasks on 3 robots. On real robots, it obtains +9%, +26%, and +28% gains on general, memory-dependent, and imagination-dependent tasks respectively.

Resources

Paper