Composite · DrainVeggies¶
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 RoboCasa365 runs for every task of a run.
Codex + GPT-6 Luna, reasoning xhigh (ChatGPT login)default runclosed
Not in this run: not in the sample (robot_coding_bench PR
Task instruction (upstream)
Dump the lettuce from the pot into the sink. Then turn on the water and wash the lettuce. Then turn off the water and put the lettuce back in the pot.
Playback speed
Recorded by us: our reference solution, replayed by the verifier (the scene camera).
What RoboCasa365 states about this task
| Success criteria | 1. The check for DrainVeggies requires that the lettuce has been washed: inside the sink with the water on for more than 10 consecutive checks (the check runs twice per control step: at least 6 control steps; once washed it stays washed); at the end the lettuce is in the pot and the water is off. 2. unlimited (privileged): both fresh-process replays of the handed-in trajectory end in the same state, and the check holds on it 3. limited (standard): the one episode (no reset) is recorded by the service and replays to the same state; the check holds on it live and in the replay |
| Family | robocasa365/DrainVeggies |
| Robot | Franka Panda on an Omron mobile base with a torso lift (PandaOmron) |
| Category | Composite · washing fruits and vegetables |
| Instance | the initial state of official demonstration episode 0 (demo_0); MuJoCo state, model arrays, task state and RNG frozen in instance.npz (SHA-256 59dae1f7b031…, checked on load) |
| Deliverable | (T, 13) native controller commands, 1 ≤ T ≤ 3000, 20 Hz |
| Reference Solution | Retargeted by PR #4 from official demonstration episode 0 (demo_0, 1250 recorded actions; RoboCasa365's demonstrations use a different arm controller): it follows the demonstration's recorded arm joint positions as this controller's absolute joint targets (PR #4 method demo_joint). Executed from the frozen scene by the robot's own controller: 1250 actions in solution/oracle.npz. Neither the demonstration nor the reference is in the image; the agent cannot reach them. |
| Limited Mode | Standard mode (robot as a service, eai-standard/2.1) of robocasa365-drain-veggies-i00-privileged: the same frozen instance and success check, served by the sim sidecar (images/robocasa365/standard/server_casa.py, tool module casa_tool), which records the episode, replays it in a fresh simulator and judges it; the verifier grades the sidecar's record in a container of its own. The agent sees the robot's cameras (RGB-D, calibrated), its own joints and end effector, and the instruction. |
| Oracle | full |
| Base Image | ghcr.io/mll-lab-nu/eai-robocasa365:0.1.0 |
| Agent Budget | 3600 s of wall clock per mode |
| Environment Source | https://github.com/robocasa/robocasa/blob/4f8a2980def75a55dff96b990745b83540425f09/robocasa/environments/kitchen/composite/washing_fruits_and_vegetables/drain_veggies.py#L4 |
| Task Dirs | robocasa365-drain-veggies-i00-privileged, robocasa365-drain-veggies-i00-standard |
From https://github.com/robocasa/robocasa @ 4f8a298 (robocasa 1.0.1), as defined in our task definitions @ 8c5594a43.
Tags¶
Task DomainManipulation
Why this task is interesting¶
Not yet written.
Capability notes¶
Not yet written.
Oracle demo review¶
Not yet reviewed.