Visualization Roadmap#

The viewer is planned for a future release. Version 0.2.0 provides link-geometry queries and serialized grasp records for external visualization. The milestones below specify the planned scope; release dates have not been assigned.

Standalone Model and Grasp Viewer#

A planned standalone viewer will present exported hands and saved grasp records through a modern three-dimensional interface. The design uses a browser-based WebGL viewport and a Python geometry adapter. Native GraspIt! remains responsible for closure and contact queries; the viewer consumes exported geometry, named joint states, and saved contact records. This separation permits inspection on a machine without the native Qt interface and preserves the existing Python planning workflow. The viewer is a future extension; the current release provides geometry queries and serialized grasp records.

The viewport will expose palm, object, and contact frames with the coordinate conventions in the theory guide. Users will inspect finger and link labels, contact positions and inward normals, and an object’s center of mass. A source/export overlay will support joint-coordinate correspondence and per-link geometric-error inspection. Mesh representation, units, model identifiers, and the active joint vector will remain visible in the model inspector.

Record Loading and Interaction#

The proposed data adapter will read the wrapper’s hand-export manifest and saved GraspCandidate records, resolve meshes within an explicitly selected asset directory, and reconstruct each named link transform. Users will reload a recorded grasp, select a candidate, inspect its contacts and quality, and adjust independent coordinates within the coupled feasible limits. A file-backed snapshot will retain the model and record hashes, camera, displayed frames, and active overlays for repeatable figure generation. Imported records remain data; loading a record will not execute embedded code or issue robot commands.

Delivery Criteria#

The first viewer milestone will cover L6, LEAP, Ability, Barrett, Shadow, and Inspire models through the same geometry adapter. Frame and normal overlays will be checked against the existing Python coordinate transforms, and record reload will reproduce the saved joint state and contact geometry. Browser rendering will be compared with native queried link geometry at the documented configurations. The implementation will provide an offscreen image-export path and keyboard-accessible controls, with no dependency on an interactive native Qt window. Hardware control and a new grasp planner remain outside this visualization extension.