Dexonomy: Synthesizing All Dexterous Grasp Types in a Grasp Taxonomy (2504.18829v1)

Abstract: Generalizable dexterous grasping with suitable grasp types is a fundamental skill for intelligent robots. Developing such skills requires a large-scale and high-quality dataset that covers numerous grasp types (i.e., at least those categorized by the GRASP taxonomy), but collecting such data is extremely challenging. Existing automatic grasp synthesis methods are often limited to specific grasp types or object categories, hindering scalability. This work proposes an efficient pipeline capable of synthesizing contact-rich, penetration-free, and physically plausible grasps for any grasp type, object, and articulated hand. Starting from a single human-annotated template for each hand and grasp type, our pipeline tackles the complicated synthesis problem with two stages: optimize the object to fit the hand template first, and then locally refine the hand to fit the object in simulation. To validate the synthesized grasps, we introduce a contact-aware control strategy that allows the hand to apply the appropriate force at each contact point to the object. Those validated grasps can also be used as new grasp templates to facilitate future synthesis. Experiments show that our method significantly outperforms previous type-unaware grasp synthesis baselines in simulation. Using our algorithm, we construct a dataset containing 10.7k objects and 9.5M grasps, covering 31 grasp types in the GRASP taxonomy. Finally, we train a type-conditional generative model that successfully performs the desired grasp type from single-view object point clouds, achieving an 82.3% success rate in real-world experiments. Project page: https://pku-epic.github.io/Dexonomy.

Overview of "Dexonomy: Synthesizing All Dexterous Grasp Types in a Grasp Taxonomy"

The paper "Dexonomy: Synthesizing All Dexterous Grasp Types in a Grasp Taxonomy" presents a comprehensive and efficient pipeline designed to synthesize dexterous grasps across varying grasp types, objects, and hand articulations. It seeks to address the challenges associated with creating high-quality, scalable datasets to advance intelligent robotic grasping skills. By leveraging an overview pipeline, the authors propose a solution that overcomes the limitations of existing methods constrained to specific grasp types or object categories.

Key Contributions

1. Novel Pipeline for Grasp Synthesis: The authors introduce a two-stage synthesis pipeline. The first stage optimizes object alignment to match a human-annotated grasp template, effectively sampling large volumes of initial poses to evade local optima issues. The second stage refines hand poses using simulation-based techniques, optimizing for realistic and penetration-free grasps.
2. Simulation-Based Validation: A notable innovation is the introduction of a contact-aware control strategy, which enables validated grasps to hold objects stably across different external force directions.
3. Creation of a Large-Scale Dataset: Using their pipeline, the researchers synthesized a dataset containing 10.7k objects and 9.5 million grasps, covering 31 of the predefined grasp types from the GRASP taxonomy, demonstrating superior scalability and quality against prior methods.
4. Generative Model and Real-World Testing: The dataset further supports training a type-conditional generative model that successfully performs grasps based on single-view object point clouds with a remarkable success rate of 82.3% in real-world experiments.

Numerical Results

The synthesis method significantly outperformed type-unaware baselines in simulation, achieving high success rates under challenging test conditions with diverse objects. The dataset offers comprehensive coverage of 31 grasp types, providing a valuable resource for further research into type-aware grasping methodologies.

Implications and Speculations

The implications of this research are profound both practically and theoretically. By effectively synthesizing a broad range of dexterous grasps, robots can engage more flexibly and intelligently with their environments, catering to task-specific requirements and improving overall utility. The methodology paves the way for more nuanced human-object interactions via robotic systems and can inspire further development in AI models that understand and replicate human-level dexterity.

Future developments might explore optimizing trajectories for dynamic grasping and tackle grasp synthesis in complex, cluttered scenes. Additionally, examining the suitability of varying grasp taxonomies and integrating trajectory planning for dynamic grasping could further enhance application scope.

Conclusion

In conclusion, the paper "Dexonomy: Synthesizing All Dexterous Grasp Types in a Grasp Taxonomy" marks a significant advancement in the field of type-aware dexterous grasp synthesis. Its contributions effectively bridge the gap between theoretical grasp synthesis and practical application, raising the standard for future research and development in intelligent robotic interaction. Such progress enhances the pursuit of robotic dexterity, setting a precedent for robust, generalized solutions in real-world applications.