Dynamic3D · Sweep into drawer¶
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 KinDER runs for every task of a run.
Codex CLI 0.157–0.159 + GPT-6 Luna, reasoning medium (OpenRouter)default run
| Trial | Started | Agent time | Model requests | Tokens in / out | Est. cost | Grade |
|---|---|---|---|---|---|---|
| U✓ 50m | 10-04 05:24 | 50m | — | 10.7M / 176k | $0.243 | deterministic 1, grader_error 0, missing_trajectory 0, n_actions 922, video_rendered 1 |
| L✗ 43m | 10-04 06:11 | 43m | — | 36.4M / 194k | $0.513 | deterministic 1, grader_error 0, live_success 0, missing_log 0, n_actions 1000, replay_success 0, video_rendered 1 |
1 other run(s) of KinDER
Codex CLI 0.159.2 + GPT-6.1 Sol, reasoning medium (OpenRouter)
| Trial | Started | Agent time | Model requests | Tokens in / out | Est. cost | Billed | Grade |
|---|---|---|---|---|---|---|---|
| U✓ 16m | 10-01 05:54 | 16m | 52 | 2.9M / 20k | $0.654 | $0.697 | deterministic 1, grader_error 0, missing_trajectory 0, n_actions 959, video_rendered 1 |
| L✓ 13m | 10-01 02:42 | 13m | 68 | 2.3M / 16k | $0.504 | $0.533 | deterministic 1, grader_error 0, live_success 1, missing_log 0, n_actions 827, replay_success 1, video_rendered 1 |
Task instruction (upstream)
Get the five small dark-red cubes on top of the kitchen island into the middle drawer of the island's upper row of drawers, which starts closed; a wiper lies next to the cubes. The task succeeds the moment all five cubes' centres are inside that drawer, each at least 2.5 cm from the drawer's two side walls.
What KinDER states about this task
| Success criteria | 1. KinDER's own goal check (the environment's terminated) fires within the episode: at most 1000 control steps (100 s at 10 Hz); no partial credit2. unlimited: both fresh-process replays of the handed-in trajectory reach it and end in the same state 3. limited: the one episode reaches it (no reset in our runs); the service records the episode, and it replays to the same state |
| Env Id | kinder/SweepIntoDrawer3D-o5-v0 |
| Robot | TidyBot++: holonomic base, Kinova Gen3 7-DoF arm, Robotiq 2F-85 gripper |
| Category | Dynamic3D (MuJoCo) |
| Instance | seed 0: the scene KinDER builds from it, pinned by its digest |
| Holding Still | does not reach the goal (checked by running an episode of zero actions) |
| Action Dim | 11 |
| Actions | base pos and yaw (3), arm joints (7), gripper pos (1); each base and joint delta at most 0.1 a step |
| State Dim | 163 |
| Limited Mode | base RGB and wrist RGB-D cameras (480×640), base odometry, the arm's joint angles and velocities, the fingers' position. Not where any object or the goal region is, not the goal check before the episode ends |
| Agent Budget | 3600 s of wall clock per mode |
From https://github.com/Princeton-Robot-Planning-and-Learning/kindergarden @ 5b2dbac, as defined in our task definitions @ ae0d92a50.
Tags¶
Why this task is interesting¶
The drawer starts closed, and its pull is a thin bar lower than the cubes on the counter: the robot has to catch the pull and open the drawer first, then get five small cubes into it (a wiper is there if it wants one). The KinDER paper's best method reaches 0.14 on this variant.
Capability notes¶
Not yet written.
Oracle demo review¶
No demo.
Discussion¶
GPT-6.1 Sol passed both modes (limited, with 50 resets: the drawer pulled open by its handle and the cubes moved in one by one; unlimited: swept in with the wiper). GPT-6 Luna (2026-10-04) passed unlimited mode with a 14 cm drawer pull and low sweeps (goal at step 922), but not the single limited episode. (@williamzhangNU)