aiqbench analysis
This research introduces ModAR, the first World-Action Model capable of autoregressively denoising multiple visual modalities—including depth maps and point tracks—before predicting robotic actions. The study demonstrates that this sequential generation approach outperforms existing formulations in efficiency and success rates while requiring significantly fewer training FLOPs than pretraining-based baselines. By systematically evaluating how different data mixtures affect performance, the work offers a more efficient pathway for developing robust multimodal agents for real-world bimanual tasks.
Modality-Autoregressive World-Action Models
arXiv:2609.17524v1 Announce Type: new Abstract: World-action models (WAMs) jointly model future observations and actions, typically predicting the future as RGB images. Other visual modalities such as depth, pretrained visual features, and point tracks can more efficiently capture geometric, semantic, and motion features. However, how best to combine these modalities within WAMs remains an open question. We introduce ModAR, the first WAM to autoregressively denoise multiple future modalities before predicting actions. This allows each prediction to condition on previously generated modalities. We train from scratch to systematically study how training-data mixtures, predicted modalities, and WAM formulations affect performance. In our evaluations, WAMs benefit from predicting point tracks, DINO features, and depth maps, while additionally predicting future RGB does not provide a consistent benefit. We also find that ModAR's sequential generation outperforms existing WAM formulations, with the highest average success rate at all evaluated data scales. We also fine-tune the video-model-initialized WAM Flex-$\pi$ on the same data; ModAR achieves a slightly higher observed average success rate (75% vs. 72%) while using approximately $20\times$ fewer training FLOPs and no pretraining. On three real-world bimanual tasks, ModAR outperforms baselines and improves with human videos.
Abstract:World-action models (WAMs) jointly model future observations and actions, typically predicting the future as RGB images. Other visual modalities such as depth, pretrained visual features, and point tracks can more efficiently capture geometric, semantic, and motion features. However, how best to combine these modalities within WAMs remains an open question. We introduce ModAR, the first WAM to autoregressively denoise multiple future modalities before predicting actions. This allows each prediction to condition on previously generated modalities. We train from scratch to systematically study how training-data mixtures, predicted modalities, and WAM formulations affect performance. In our evaluations, WAMs benefit from predicting point tracks, DINO features, and depth maps, while additionally predicting future RGB does not provide a consistent benefit. We also find that ModAR's sequential generation outperforms existing WAM formulations, with the highest average success rate at all evaluated data scales. We also fine-tune the video-model-initialized WAM Flex-$\pi$ on the same data; ModAR achieves a slightly higher observed average success rate (75% vs. 72%) while using approximately $20\times$ fewer training FLOPs and no pretraining. On three real-world bimanual tasks, ModAR outperforms baselines and improves with human videos.
Robotics (cs.RO)
arXiv:2609.17524 [cs.RO]
(or arXiv:2609.17524v1 [cs.RO] for this version)
https://doi.org/10.48550/arXiv.2609.17524 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 278 words
Related coverage
Runway’s WorldPrompt and the Engineering of Real-Time Worlds
Latent Space · Sep 25, 2026
SUAVE: Unified Video-Action Models via Masked Diffusion
arXiv · Oct 6, 2026
Pelican-Sim 1.0: A General World Model Simulator for Embodied Intelligence
arXiv · Sep 14, 2026
Build a Physical AI model factory with NVIDIA Cosmos 3 on SageMaker HyperPod
AWS · Sep 4, 2026