GET Asymmetric Gripper Grasp Planners: Dual Breakthroughs in 2D and 3D Pathways

Asymmetric Grippers Get Dedicated Grasp Planning Solutions

In the field of robotic grasping, most grasp planning algorithms are designed for symmetric parallel-jaw grippers, yet an increasing number of real-world applications require asymmetric grippers to handle complex object manipulation tasks. A recent paper published on arXiv introduces two grasp planning methods specifically designed for GET asymmetric grippers — GET-2D-1.0 and GET-3D-1.0 — approaching the problem from 2D perception and 3D modeling pathways respectively, offering entirely new solutions for efficient grasping with asymmetric grippers.

Two Technical Pathways: Rapid Perception and Precision Modeling

GET-2D-1.0: Fast Planning Based on Single-View RGB-D Images

GET-2D-1.0 is a fast grasp planner that requires only a single-view RGB-D image to compute grasp poses. The method employs the classic Ferrari-Canny metric to evaluate grasp quality and introduces a novel sampling strategy that efficiently searches for optimal grasp poses under limited perceptual information. This design makes it particularly suitable for applications with high real-time requirements, such as assembly line sorting and warehouse logistics.

GET-3D-1.0: Fine-Grained Planning Based on Mesh Models

GET-3D-1.0 takes a more refined approach. This method leverages the complete 3D mesh model of the gripper and combines Ray-Tracing technology to analyze the geometric relationship between the gripper and target objects. By precisely simulating the contact between the gripper and object surfaces during the closing process, GET-3D-1.0 can plan more stable and reliable grasps in three-dimensional space, making it especially suitable for scenarios demanding extremely high grasping precision.

Physical Experiment Validation: Over 40% Improvement in Lift Success Rate

The research team went beyond simulation testing, validating the practical performance of both planners through physical experiments with real robots. Results show that GET-2D-1.0 achieved a significant improvement of over 40% in object lift success rate compared to the traditional Bounding Box baseline method. This result clearly demonstrates that dedicated planning algorithms for asymmetric grippers hold a distinct advantage over general-purpose methods.

Notably, the research team conducted systematic comparisons between both methods and multiple baselines, further validating the robustness and generalization capabilities of the proposed solutions across different object types and grasping scenarios.

Technical Significance and Future Outlook

The core contribution of this research lies in breaking the long-standing "symmetric gripper-centric" paradigm in grasp planning. As robotic end-effector designs become increasingly diverse — with asymmetric, multi-fingered, and soft grippers continually emerging — traditional general-purpose grasp planning methods can no longer fully leverage the capabilities of these novel grippers.

From a practical standpoint, GET-2D-1.0's fast inference capability gives it strong potential for industrial deployment, while GET-3D-1.0's high-precision modeling opens possibilities for advanced manufacturing scenarios such as precision assembly. The two methods complement each other, allowing users to flexibly balance speed and accuracy based on their specific application requirements.

Looking ahead, as deep learning and geometric reasoning continue to converge, grasp planning for asymmetric grippers is expected to achieve stronger generalization in increasingly complex cluttered environments, driving robotic manipulation from merely "being able to grasp" to truly "knowing how to grasp."