DreamVLA: A Vision-Language-Action Model Dreamed with Comprehensive World Knowledge

Xin Jin (Eastern Institute of Technology, Ningbo) · Zekun Qi (Tsinghua University) · Wenyao Zhang (Shanghai Jiaotong University) · Runpei Dong (University of Illinois Urbana-Champaign) · XinQiang Yu (Galbot) · Jiazhao Zhang (Peking University) · Jiawei He · Zhizheng Zhang (University of Science and Technology of China) · He Wang (Peking University) · Li Yi (Tsinghua University, Tsinghua University) · Hongsi Liu (Eastern Institute of Technology) · Yunnan Wang (Shanghai Jiao Tong University) · Wenjun Zeng (Eastern Institute of Technology, Ningbo)
action predictionaction representationsaction-forecasting loopcalvin abc-d benchmarksdiffusion-based transformerdynamic-region-guided predictiongeneralizationimage generationinverse dynamics modelingmultimodal reasoningmultimodal uncertaintyreasoningrobot manipulationvision-language-actionworld knowledge forecasting

Recent advances in vision-language-action (VLA) models have shown promise in integrating image generation with action prediction to improve generalization and reasoning in robot manipulation. However, existing methods are limited to challenging image-based forecasting, which suffers from redundant information and lacks comprehensive and critical world knowledge, including geometry, semantics and spatial information. To address these limitations, we propose DreamVLA, a novel VLA framework that integrates comprehensive world knowledge forecasting to enable inverse dynamics modeling, thereby establishing an action-forecasting loop for manipulation tasks. Specifically, DreamVLA introduces a dynamic-region-guided world knowledge prediction mechanism, which anticipates visual, depth, geometric, semantic, and segmentation cues to provide compact yet comprehensive representations for action planning. This design aligns with how humans interact with the world by first forming abstract multimodal reasoning chains before acting. Moreover, to model the conditional distribution over future actions, we employ a diffusion-based transformer that disentangles action representations from shared latent features and better captures multimodal uncertainty. Extensive experiments on both real-world and simulation environments demonstrate that DreamVLA achieves 76.7% success rate on real robot tasks and 4.45 average length on the CALVIN ABC-D benchmarks.