HiveBoard is a fully 3D-printed, open benchmark for evaluating grippers, hands, and manipulation policies on functional mechanisms โ valves, locks, drawers, switches, threaded fasteners โ instead of loose, free-floating objects.
A hexagonal honeycomb base accepts 12 interchangeable attachments, grouped into three skill categories:
| Category | Examples |
|---|---|
| ๐ง Torque-based | Ball valve, gate valves, circuit breaker, light bulb |
| ๐ฏ Precision-based | Threaded fasteners, peg insertion, button |
| ๐งฉ Composed assembly | Lock & key, drawer, shock absorber |
Every attachment prints on a consumer-grade FDM printer and ships with articulated URDF/USD assets for direct use in simulators like MuJoCo and Isaac Sim.
Manipulation research keeps benchmarking on grasping loose objects โ HiveBoard is built to cross-validate real embodiments (arms, legged manipulators, prosthetic hands) on the same functional tasks, so results are comparable across labs, not just within one.
We're growing a shared, cross-laboratory evaluation dataset. Contributing is simple:
Evaluation data submitted to this project is released under CC BY 4.0. See the registration form for full licensing and privacy terms.
Open an issue on our GitHub or reach out to ricardo.godoy@usp.br.