A contact task in three dimensions.
UR5e + Robotiq 2F-85 · Free rigid block, table and recorded task attempts.
Replay excerpt, complete denominator.
9 recorded attempts from 3 scenes are shown here. The complete benchmark contains 18 planned attempts across 6 scenes; this excerpt is not a success-rate denominator. All source attempts, including invalid attempts and negative controls, remain in the linked receipt.
Targets are issued at the recorded observation boundary. At the terminal row, the viewer carries forward the last issued target for reference; no new terminal command is invented. Targets are requested controller commands, not measured joint torque or fingertip force.
| Scene | Controller | Recorded outcome | Reason | Source |
|---|---|---|---|---|
| nominal-00 | placement | Completed | success | Exact attempt JSON |
| nominal-00 | replay_open_jaw | Failed | timeout | Exact attempt JSON |
| nominal-00 | hold | Failed | timeout | Exact attempt JSON |
| low_friction-00 | placement | Completed | success | Exact attempt JSON |
| low_friction-00 | replay_open_jaw | Failed | timeout | Exact attempt JSON |
| low_friction-00 | hold | Failed | timeout | Exact attempt JSON |
| nominal-03 | placement | Completed | success | Exact attempt JSON |
| nominal-03 | replay_open_jaw | Failed | timeout | Exact attempt JSON |
| nominal-03 | hold | Failed | timeout | Exact attempt JSON |
9 source attempts are outside the render selection or numerically invalid. Their exact IDs and exclusion reasons are listed in motion.json. The renderer never scores task success.
Six degrees of freedom, three-dimensional physics.
Six independently commanded arm joints move and orient the tool. Its pose has three position coordinates (X, Y, Z) and three rotational degrees of freedom. The mechanism and contacts are simulated in 3D; this viewer displays the recorded result. The browser does not rerun physics.
These are authored-controller task attempts with a free rigid block and table. Success and failure labels come from the sealed source measurements. They do not demonstrate a learned policy, shoe manipulation or physical robot qualification. The grid, axes and joint markers are viewing aids.
Inspect the evidence.
Visual mode shows the model’s native surfaces. Collision mode shows the simpler contact proxies used by the simulator. The upstream fixed base has no collision proxy.