Asynchronous Fast-Slow Vision-Language-Action Policies for Whole-Body Robotic Manipulation

2025-12-23

AI Summary

Current Vision-Language-Action (VLA) systems enforce synchronous execution between vision-language models (VLMs) and action modules, severely limiting real-time performance and control stability for whole-body robotic manipulation due to large-model inference latency. To address this, we propose DuoCore-FS—the first asynchronous dual-path VLA framework featuring fast and slow processing streams. It introduces a latent-variable buffer and a whole-body action tokenizer to enable end-to-end decoupled training of semantic understanding (using a 3B-parameter VLM) and high-frequency action generation (at 30 Hz). This architecture allows low-frequency semantic updates from the VLM to coexist with high-frequency responses from a dedicated action expert, effectively increasing control bandwidth by 3×. In real-robot experiments, DuoCore-FS achieves significantly higher task success rates and faster response times compared to synchronous baselines.

Technology Category

• Computer Vision: Language and Vision
• Intelligent Robots: Multimodal Perception & Sensor Fusion
• Machine Learning: Large Multimodal Models (LMMs)

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications
• Search and Retrieval-Augmented AI: Multilingual and cross-lingual Web search

Abstract

Most Vision-Language-Action (VLA) systems integrate a Vision-Language Model (VLM) for semantic reasoning with an action expert generating continuous action signals, yet both typically run at a single unified frequency. As a result, policy performance is constrained by the low inference speed of large VLMs. This mandatory synchronous execution severely limits control stability and real-time performance in whole-body robotic manipulation, which involves more joints, larger motion spaces, and dynamically changing views. We introduce a truly asynchronous Fast-Slow VLA framework (DuoCore-FS), organizing the system into a fast pathway for high-frequency action generation and a slow pathway for rich VLM reasoning. The system is characterized by two key features. First, a latent representation buffer bridges the slow and fast systems. It stores instruction semantics and action-reasoning representation aligned with the scene-instruction context, providing high-level guidance to the fast pathway. Second, a whole-body action tokenizer provides a compact, unified representation of whole-body actions. Importantly, the VLM and action expert are still jointly trained end-to-end, preserving unified policy learning while enabling asynchronous execution. DuoCore-FS supports a 3B-parameter VLM while achieving 30 Hz whole-body action-chunk generation, approximately three times as fast as prior VLA models with comparable model sizes. Real-world whole-body manipulation experiments demonstrate improved task success rates and significantly enhanced responsiveness compared to synchronous Fast-Slow VLA baselines. The implementation of DuoCore-FS, including training, inference, and deployment, is provided to commercial users by Astribot as part of the Astribot robotic platform.

Problem

• Asynchronous framework improves robotic manipulation speed
• Separates fast action from slow semantic reasoning
• Enhances real-time performance and control stability

Innovation

• Asynchronous fast-slow pathways for high-frequency action generation
• Latent representation buffer bridges slow semantic reasoning and fast execution
• Whole-body action tokenizer provides compact unified representation for manipulation