GE-Act 2.0: Pretraining and Scaling a World-Action Model for Robotic Manipulation

Hugging Face Daily Papers Papers

Summary

This paper introduces GE-Act 2.0, a world-action model pretrained from scratch to enable scalable zero-shot robotic manipulation with improved success rates across diverse tasks and conditions.

World-action models (WAM) predict future states to guide robot actions, enabling learning from both action-free video and action-labeled interaction. Most inherit pretrained video generators, leaving WAM pretraining and scaling underexplored. We introduce Genie Envisioner Act 2.0 (GE-Act 2.0), a world-action model whose trainable generative and action components are all initialized from scratch on manipulation data. It combines a control-oriented autoencoder (CoAE), a single-step visual planner (SVP), and an inverse dynamics model (IDM). CoAE retains action- and instruction-relevant information under aggressive compression, while SVP produces a complete future state in one differentiable pass, so visual planning and inverse dynamics can be pretrained separately on complementary data. The components are then jointly trained with knowledge-aligned selective optimization (KASO), which reduces mismatched supervision by selecting only predicted futures judged behaviorally compatible with the recorded action. We evaluate pretrained checkpoints directly, without per-task fine-tuning, on 100 tasks across 20 manipulation skill groups with held-out scenes, backgrounds, lighting, and object instances. Scaling co-training data from 300 to 30,000 hours raises success from 17.1% to 44.1% on G1-OP and from 13.4% to 31.1% on G2-90D; despite comprising less than 2% of the co-training data, G2-90D improves by 17.7 points, suggesting cross-embodiment transfer. Gains span 19/20 and 18/20 skill groups, and skill-specific coverage strongly correlates with zero-shot out-of-distribution (OOD) success (Pearson r=0.80; Spearman rho=0.85). Under the same protocol, the model grounds object, color, shape, and position references in at least 90% of trials and follows explicit instructions even when they conflict with an already-committed behavior or a conventional scene association.
Original Article
View Cached Full Text

Cached at: 09/09/26, 04:29 AM

Paper page - GE-Act 2.0: Pretraining and Scaling a World-Action Model for Robotic Manipulation

Source: https://huggingface.co/papers/2609.05588 Authors:

,

,

,

,

,

,

,

,

,

,

,

,

,

,

,

,

,

,

,

,

Abstract

GE-Act 2.0 is a world-action model trained from scratch with a control-oriented autoencoder, single-step visual planner, and inverse dynamics model, using knowledge-aligned selective optimization to enable scalable zero-shot robot manipulation across diverse skills and conditions.

World-action models(WAM) predict future states to guide robot actions, enabling learning from both action-free video and action-labeled interaction. Most inherit pretrained video generators, leaving WAM pretraining and scaling underexplored. We introduce Genie Envisioner Act 2.0 (GE-Act 2.0), a world-action model whose trainable generative and action components are all initialized from scratch on manipulation data. It combines acontrol-oriented autoencoder(CoAE), asingle-step visual planner(SVP), and aninverse dynamics model(IDM). CoAE retains action- and instruction-relevant information under aggressive compression, while SVP produces a complete future state in one differentiable pass, so visual planning and inverse dynamics can be pretrained separately on complementary data. The components are then jointly trained withknowledge-aligned selective optimization(KASO), which reduces mismatched supervision by selecting only predicted futures judged behaviorally compatible with the recorded action. We evaluate pretrained checkpoints directly, without per-task fine-tuning, on 100 tasks across 20 manipulation skill groups with held-out scenes, backgrounds, lighting, and object instances. Scaling co-training data from 300 to 30,000 hours raises success from 17.1% to 44.1% on G1-OP and from 13.4% to 31.1% on G2-90D; despite comprising less than 2% of the co-training data, G2-90D improves by 17.7 points, suggestingcross-embodiment transfer. Gains span 19/20 and 18/20 skill groups, and skill-specific coverage strongly correlates withzero-shot out-of-distribution(OOD) success (Pearson r=0.80; Spearman rho=0.85). Under the same protocol, the model grounds object, color, shape, and position references in at least 90% of trials and follows explicit instructions even when they conflict with an already-committed behavior or a conventional scene association.

View arXiv pageView PDFProject pageAdd to collection

Get this paper in your agent:

hf papers read 2609\.05588

Don’t have the latest CLI?curl \-LsSf https://hf\.co/cli/install\.sh \| bash

Models citing this paper0

No model linking this paper

Cite arxiv.org/abs/2609.05588 in a model README.md to link it from this page.

Datasets citing this paper0

No dataset linking this paper

Cite arxiv.org/abs/2609.05588 in a dataset README.md to link it from this page.

Spaces citing this paper0

No Space linking this paper

Cite arxiv.org/abs/2609.05588 in a Space README.md to link it from this page.

Collections including this paper0

No Collection including this paper

Add this paper to acollectionto link it from this page.

Similar Articles

Geometric Action Model for Robot Policy Learning

Hugging Face Daily Papers

The Geometric Action Model (GAM) repurposes a pretrained geometric foundation model (GFM) as a unified backbone for language-conditioned robot manipulation, achieving higher accuracy, robustness, and efficiency than existing foundation-model-scale baselines across simulation and real-world benchmarks.

ABot-M0.5: Unified Mobility-and-Manipulation World Action Model

Hugging Face Daily Papers

ABot-M0.5 is a new World Action Model for mobile manipulation that improves performance through temporal granularity alignment, action space disentanglement, and train-test consistency, achieving state-of-the-art results on long-horizon and fine-grained manipulation benchmarks.

Learning Transferable Dynamics Priors from Action to World Modeling

Hugging Face Daily Papers

This paper introduces A2World, a diffusion-based world model pretrained on large-scale robot manipulation data to learn transferable dynamics priors. The model can be adapted into a real-world simulator (A2World-sim) for policy evaluation or a video-action prediction model (A2World-policy) for action prediction, demonstrating benefits for both simulator-centric and policy-centric robot learning.