MachaGrasp: Morphology-Aware Cross-Embodiment Dexterous Hand Articulation Generation for Grasping
Heng Zhang, Kevin Yuchen Ma, Zheng SHOU, Weisi Lin, Yan Wu
AI summary
Problem
Existing dexterous grasp methods are typically hand-specific, requiring large datasets and retraining for each new embodiment, while optimization-based cross-embodiment approaches are computationally expensive and slow.
Approach
The framework extracts morphology embeddings and eigengrasp bases directly from a hand's URDF, then uses a neural amplitude predictor to regress low-dimensional articulation coefficients conditioned on object point clouds and wrist pose, which are decoded into full joint configurations.
Key results
- 91.9% average grasp success rate across three unseen hands in simulation
- <0.4 seconds inference time per grasp
- 85.6% simulation and 87% real-world success with few-shot adaptation to a new hand
- Outperforms optimization-based and prior learning-based baselines in accuracy and efficiency
Why it matters
Provides a scalable, end-to-end solution for deploying dexterous grasping policies across diverse robotic hardware without per-hand data collection or retraining.
Abstract
Dexterous grasping with multi-fingered hands re- mains challenging due to high-dimensional articulations and the cost of optimization-based pipelines. Existing end-to-end meth- ods require training on large-scale datasets for specific hands, limiting their ability to generalize across different embodiments. We propose MachaGrasp, an eigengrasp-based, end-to-end framework for cross-embodiment grasp generation. From a hand’s morphology description, we derive a morphology embed- ding and an eigengrasp set. Conditioned on these, together with the object point cloud and wrist pose, an amplitude predictor regresses articulation coefficients in a low-dimensional space, which are decoded into full joint articulations. Articulation learning is supervised with a Kinematic-Aware Articulation Loss (KAL) that emphasizes fingertip-relevant motions and injects morphology-specific structure. In simulation on unseen objects across three dexterous hands, MachaGrasp attains a 91.9% average grasp success rate with <0.4 s inference per grasp. With few-shot adaptation to an unseen hand, it achieves 85.6% success on unseen objects in simulation, and real-world experiments on this few-shot-generalized hand achieve an 87% success rate. The code and additional materials are available on our project website https://connor-zh.github.io/MachaGrasp/.