Retrieve Cube¶
Agent runs
Each mode of this task runs once per run. Est. cost is tokens at list price (data/prices.yml), never a bill; see the RoboWits runs for every task of a run.
Codex + GPT-6 Luna, reasoning xhigh (ChatGPT login)default run
| Trial | Started | Agent time | Model requests | Tokens in / out | Est. cost | Grade |
|---|---|---|---|---|---|---|
| U✓ 57m | 09-26 02:28 | 57m | 113 | 11.0M / 101k | $0.181 | deterministic 1, grader_error 0, instance_exact 1, instance_ok 1, n_actions 1240, video_rendered 1 |
| L✗ 1h 30m | 09-29 13:00 | 1h 30m | 154 | 19.4M / 140k | $0.294 | deterministic 1, grader_error 0, live_success 0, n_actions 2340, replay_success 0, video_rendered 1 |
Task instruction (upstream)
Retrieve the cube from the container and place it on the green target area.
Playback speed
What RoboWits states about this task
| Success criteria | 1. Cube is on the table surface (z within tolerance of TABLE_Z) 2. Cube is outside the container (2D footprint overlap with container < 50%) 3. Cube's 2D footprint overlaps target area by at least 50% 4. Objects remain within table bounds |
| Objects | cube deep narrow container target area |
| Physics | rigid |
| Episode budget | 200 control steps |
| Default control mode | EE_ABS |
| Evaluation scenes | 50 |
| Published mutations | 5 |
| BlenderKit assets | 1 (not redistributable — mounted from the host) |
| Registry id | robowits/02-retrieve-cube-v0 |
| Environment class | RetrieveCubeEnv |
| Defined in | gs_gym/envs/robowits/02_retrieve_cube.py |
From https://github.com/UMass-Embodied-AGI/RoboWits @ 9cc30ae.
Tags¶
Task DomainManipulation
Why this task is interesting¶
Not yet written.
Capability notes¶
Not yet written.
Oracle demo review¶
Not yet reviewed.