EAGG: Embodiment-Aligned Grasp Generation via Geometry-Aware Graph Conditioning

2026-06-16 · Source: Artificial Intelligence · Field: Technology & Digital — Robotics & Autonomous Systems, Artificial Intelligence & Machine Learning · Depth: Expert, quick

Summary

EAGG, an embodiment-aligned grasp generator, addresses the challenge of creating a unified model for robotic end effectors ranging from parallel grippers to dexterous hands. Traditional grasp generators often fail to transfer effectively across diverse gripper topologies, actuation, and contact geometries. EAGG overcomes this by representing each embodiment with a topology-aware end-effector graph and an embodiment-specific low-dimensional control space. It utilizes a frozen end-effector-cognition backbone to generate geometry-aware tokens, which serve as a reusable morphology prior. Crucially, iterative geometry injection continuously updates these tokens during sampling, ensuring conditioning remains synchronized with the evolving end-effector geometry. On the MultiGripperGrasp benchmark, EAGG achieved a 56.17% average success rate across six training end effectors, closely matching specialized training performance within 1.10 percentage points. This approach also improved contact distance, reducing the pooled median from 0.239 cm to 0.189 cm.

Key takeaway

For robotics engineers developing universal grasp generation systems, EAGG demonstrates that explicitly aligning embodiment structure within a shared model significantly improves performance and transferability. You should consider integrating topology-aware end-effector graphs and iterative geometry injection into your models to achieve robust cross-end-effector generalization. This approach can yield high success rates, like EAGG's 56.17% on MultiGripperGrasp, while maintaining adaptability to new gripper designs.

Key insights

Aligning embodiment structure within a shared generator strengthens cross-end-effector grasp generation.

Principles

• Represent embodiments with topology-aware graphs.
• Use embodiment-specific control spaces.
• Continuously update geometry-aware tokens.

Method

EAGG uses a frozen end-effector-cognition backbone to create geometry-aware tokens, then iteratively injects geometry to synchronize conditioning with evolving end-effector states during grasp sampling.

In practice

• Achieve 56.17% grasp success across diverse grippers.
• Reduce median contact distance to 0.189 cm.
• Preserve transfer to zero-shot end effectors.

Topics

• Grasp Generation
• Robotic End Effectors
• Embodiment Alignment
• MultiGripperGrasp
• Zero-Shot Transfer

Code references

• wanhaoniu/EAGG

Best for: Research Scientist, AI Scientist, Robotics Engineer, Machine Learning Engineer

Related on AIssential

• EquiDexFlow: Contact-Grounded SE(3)-Equivariant Dexterous Grasp Generative Flows — Jointly predicting grasp kinematics and contact forces ensures physically stable dexterous grasps.
• ABot-M0: VLA Foundation Model for Robotic Manipulation with Action Manifold Learning — Unified data, action manifold learning, and modular 3D perception enable generalizable robotic manipulation across diverse hardware.
• Rostok Framework: Automating Underactuated Robot Gripper Design — Combining morphological computation and generative design automates the creation of effective robot grippers.
• A Cross-view Fusion Framework for Robust 6-DoF Grasp Pose Estimation — Cross-view fusion with self-supervised contrastive learning and cylindrical integration robustly enhances 6-DoF grasp estimation in occluded scenes.
• Grasp-Then-Plan with Failure Attribution: A Closed Two-Stage Framework for Precise and Generalizable Robotic Manipulation — GTP-FA improves robotic manipulation by attributing failures to optimize distinct grasping and planning stages.
• Blind Dexterous Grasping via Real2Sim2Real Tactile Policy Learning — Combining calibrated simulation, geometry-aware tactile encoding, and diffusion policies enables blind dexterous grasping.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Artificial Intelligence.