Proposal
Expand the built-in parallel-jaw gripper geometry catalog exposed by get_parallel_jaw_gripper_model() beyond the initial dh_pgi_140_80 preset.
Candidate families include Robotiq 2F-85 / 2F-140, Franka Hand, OnRobot RG2 / RG6, and Weiss WSG-50. The final list should follow actual task reuse and availability of authoritative geometry data.
Each preset should contain only grasp-planning geometry consumed by GraspKit, such as the model ID, opening limits, finger dimensions, and palm depth. URDF paths, downloadable assets, robot registration, and runtime-mutable plugin state should remain outside this catalog.
Motivation
Reusable model IDs keep task Gym configs concise and prevent the same physical dimensions from being copied, drifting, or being interpreted differently across Expert Program environments. Inline ParallelJawGripperModelCfg values remain useful for custom or calibrated hardware, while common production grippers should have reviewed toolkit-owned presets.
Additional context
Acceptance criteria for each added preset:
- verify dimensions against manufacturer documentation or an authoritative, versioned model and record the source;
- use metres and the existing GraspKit frame/field semantics consistently;
- distinguish hardware or finger variants with stable model IDs when their planning geometry differs;
- return a fresh independent
ParallelJawGripperModelCfg from each lookup;
- add focused value, validation, independence, and API documentation coverage;
- avoid a public mutable global registration API unless a separate extensibility use case is established.
The initial dh_pgi_140_80 preset establishes the intended static-catalog pattern in embodichain/toolkits/graspkit/pose_generator.py.
Checklist
Proposal
Expand the built-in parallel-jaw gripper geometry catalog exposed by
get_parallel_jaw_gripper_model()beyond the initialdh_pgi_140_80preset.Candidate families include Robotiq 2F-85 / 2F-140, Franka Hand, OnRobot RG2 / RG6, and Weiss WSG-50. The final list should follow actual task reuse and availability of authoritative geometry data.
Each preset should contain only grasp-planning geometry consumed by GraspKit, such as the model ID, opening limits, finger dimensions, and palm depth. URDF paths, downloadable assets, robot registration, and runtime-mutable plugin state should remain outside this catalog.
Motivation
Reusable model IDs keep task Gym configs concise and prevent the same physical dimensions from being copied, drifting, or being interpreted differently across Expert Program environments. Inline
ParallelJawGripperModelCfgvalues remain useful for custom or calibrated hardware, while common production grippers should have reviewed toolkit-owned presets.Additional context
Acceptance criteria for each added preset:
ParallelJawGripperModelCfgfrom each lookup;The initial
dh_pgi_140_80preset establishes the intended static-catalog pattern inembodichain/toolkits/graspkit/pose_generator.py.Checklist