DynamicVLA: A Vision-Language-Action Model for Dynamic Object Manipulation

2026-01-29

AI Summary

This work addresses the challenge of simultaneously achieving rapid perception, temporal prediction, and continuous control in dynamic object manipulation with existing vision–language–action (VLA) models. To this end, we propose DynamicVLA, a novel framework featuring a compact 0.4B-parameter convolutional visual encoder, a continuous inference mechanism, and a latent-variable-aligned action stream to enable efficient closed-loop control. We further introduce DOM, the first large-scale dataset for dynamic manipulation, along with an automated synthetic data generation pipeline that facilitates cross-entity generalization. Experimental results demonstrate that our approach significantly improves response latency, dynamic scene understanding, and generalization capability, achieving state-of-the-art performance in both simulated and real-world dynamic manipulation tasks.

Technology Category

• Computer Vision: Large Vision Models
• Intelligent Robots: Manipulation
• Humans and AI: Human-Aware Planning and Behavior Prediction

Application Category

• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs
• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications
• Responsible Web: Machine-in-the-loop, human agency and autonomy

Abstract

Manipulating dynamic objects remains an open challenge for Vision-Language-Action (VLA) models, which, despite strong generalization in static manipulation, struggle in dynamic scenarios requiring rapid perception, temporal anticipation, and continuous control. We present DynamicVLA, a framework for dynamic object manipulation that integrates temporal reasoning and closed-loop adaptation through three key designs: 1) a compact 0.4B VLA using a convolutional vision encoder for spatially efficient, structurally faithful encoding, enabling fast multimodal inference; 2) Continuous Inference, enabling overlapping reasoning and execution for lower latency and timely adaptation to object motion; and 3) Latent-aware Action Streaming, which bridges the perception-execution gap by enforcing temporally aligned action execution. To fill the missing foundation of dynamic manipulation data, we introduce the Dynamic Object Manipulation (DOM) benchmark, built from scratch with an auto data collection pipeline that efficiently gathers 200K synthetic episodes across 2.8K scenes and 206 objects, and enables fast collection of 2K real-world episodes without teleoperation. Extensive evaluations demonstrate remarkable improvements in response speed, perception, and generalization, positioning DynamicVLA as a unified framework for general dynamic object manipulation across embodiments.

Problem

• dynamic object manipulation
• Vision-Language-Action models
• temporal anticipation
• continuous control
• perception-execution gap

Innovation

• DynamicVLA
• Continuous Inference
• Latent-aware Action Streaming
• Vision-Language-Action Model
• Dynamic Object Manipulation