Global Prior Meets Local Consistency: Dual-Memory Augmented Vision-Language-Action Model for Efficient Robotic Manipulation

2026-02-22

AI Summary

This work addresses the inefficiency and poor robustness of existing vision-language-action (VLA) models, which stem from their reliance on isotropic Gaussian noise priors and neglect of temporal consistency in action sequences. To overcome these limitations, we propose OptimusVLA, the first VLA framework incorporating a dual-memory mechanism: a Global Prior Memory (GPM) that retrieves semantically similar trajectories to provide task-level priors—replacing Gaussian noise and accelerating generation—and a Local Consistency Memory (LCM) that models executed action sequences to enhance temporal coherence and trajectory smoothness. OptimusVLA achieves state-of-the-art performance with a 98.6% success rate on LIBERO, outperforming pi₀ by 13.5% on CALVIN and by 38% on RoboTwin 2.0 Hard. It also demonstrates a 42.9% improvement in real-world generalization and a 52.4% gain in long-horizon task success, while achieving a 2.9× speedup in inference.

Technology Category

• Computer Vision: Language and Vision
• Search and Optimization: Learning to Search
• Intelligent Robots: Learning & Optimization for ROB

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Search and Retrieval-Augmented AI: Large language models for search
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Abstract

Hierarchical Vision-Language-Action (VLA) models have rapidly become a dominant paradigm for robotic manipulation. It typically comprising a Vision-Language backbone for perception and understanding, together with a generative policy for action generation. However, its performance is increasingly bottlenecked by the action generation proceess. (i) Low inference efficiency. A pronounced distributional gap between isotropic noise priors and target action distributions, which increases denoising steps and the incidence of infeasible samples. (ii) Poor robustness. Existing policies condition solely on the current observation, neglecting the constraint of history sequence and thus lacking awareness of task progress and temporal consistency. To address these issues, we introduce OptimusVLA, a dual-memory VLA framework with Global Prior Memory (GPM) and Local Consistency Memory (LCM). GPM replaces Gaussian noise with task-level priors retrieved from semantically similar trajectories, thereby shortening the generative path and reducing the umber of function evaluations (NFE). LCM dynamically models executed action sequence to infer task progress and injects a learned consistency constraint that enforces temporal coherence and smoothness of trajectory. Across three simulation benchmarks, OptimusVLA consistently outperforms strong baselines: it achieves 98.6% average success rate on LIBERO, improves over pi_0 by 13.5% on CALVIN, and attains 38% average success rate on RoboTwin 2.0 Hard. In Real-World evaluation, OptimusVLA ranks best on Generalization and Long-horizon suites, surpassing pi_0 by 42.9% and 52.4%, respectively, while delivering 2.9x inference speedup.

Problem

• Vision-Language-Action
• action generation
• inference efficiency
• temporal consistency
• robustness

Innovation

• Dual-Memory
• Global Prior Memory
• Local Consistency Memory
• Vision-Language-Action Model
• Efficient Robotic Manipulation