Dex1B: Learning with 1B Demonstrations for Dexterous Manipulation

2025-06-20

AI Summary

Generating large-scale, high-quality demonstration data for dexterous hand manipulation faces a fundamental trade-off between task diversity and physical feasibility. To address this, we propose the first billion-scale dexterous manipulation demonstration dataset, covering two foundational tasks: grasping and joint actuation. We design a generative framework that jointly incorporates geometric constraint modeling and conditional diffusion mechanisms, enabling controllable synthesis of demonstrations with both high physical feasibility and rich task diversity. Furthermore, we introduce a simulation-to-real transfer technique to enhance deployment robustness. Our approach achieves significant improvements over state-of-the-art methods across multiple simulated benchmarks and successfully enables zero-shot transfer control on real dexterous hands—demonstrating strong generalization and practical applicability.

Technology Category

• Intelligent Robots: Manipulation
• Search and Optimization: Sampling/Simulation-based Search
• Computer Vision: Diffusion Models for Vision

Application Category

• Economics, Online Markets and Human Computation: Economic ramifications for generative AI infrastructure and applications
• Web Mining and Content Analysis: Web data generation and simulation
• User Modeling, Personalization and Recommendation: User modeling and simulation for interactive and conversational systems

Abstract

Generating large-scale demonstrations for dexterous hand manipulation remains challenging, and several approaches have been proposed in recent years to address this. Among them, generative models have emerged as a promising paradigm, enabling the efficient creation of diverse and physically plausible demonstrations. In this paper, we introduce Dex1B, a large-scale, diverse, and high-quality demonstration dataset produced with generative models. The dataset contains one billion demonstrations for two fundamental tasks: grasping and articulation. To construct it, we propose a generative model that integrates geometric constraints to improve feasibility and applies additional conditions to enhance diversity. We validate the model on both established and newly introduced simulation benchmarks, where it significantly outperforms prior state-of-the-art methods. Furthermore, we demonstrate its effectiveness and robustness through real-world robot experiments. Our project page is at https://jianglongye.com/dex1b

Problem

• Generating large-scale dexterous hand manipulation demonstrations
• Improving feasibility and diversity in generative models
• Validating performance on simulation and real-world benchmarks

Innovation

• Generative models create diverse demonstrations
• Integrates geometric constraints for feasibility
• Applies conditions to enhance diversity