Paper 2601.22153
DynamicVLA: A Vision-Language-Action Model for Dynamic Object Manipulation
- Published
- Jan 2026
- Research lab
- Independent
- Citations
- 24
- GitHub
- 318 stars
01 In brief
Summary
DynamicVLA is a 0.4B-parameter Vision-Language-Action (VLA) model designed for dynamic object manipulation, addressing latency issues in existing VLAs.
It introduces three key innovations: a compact architecture using a convolutional vision encoder (FastViT) and a truncated SmolLM2-360M backbone for fast inference; Continuous Inference, which overlaps reasoning and execution to eliminate inter-chunk waiting; and Latent-aware Action Streaming (LAAS), which discards outdated actions and prioritizes recent predictions to maintain temporal alignment.
The authors also introduce the Dynamic Object Manipulation (DOM) benchmark, with 200K synthetic episodes across 2.8K scenes and 206 objects, plus 2K real-world episodes collected via an automated pipeline without teleoperation.
In simulation, DynamicVLA achieves 47.06% average success rate, outperforming baselines like SmolVLA (12.67%) and VLASH (12.33%).
Real-world tests on Franka and PiPER arms show consistent improvements in interaction, perception, and generalization tasks.
Ablations confirm the importance of the 360M backbone, FastViT, Continuous Inference, and LAAS.
The work highlights that temporal misalignment, not perceptual ambiguity, is the main failure mode in dynamic manipulation.
02 From the paper
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.