Manipulation · Cabinet¶
Excluded from the benchmark
its bar cannot be reached at upstream's commit (robot_coding_bench 2026-10-04)
Greyed and last in the task list, left out of the benchmark's numbers; its demo and agent runs stay (excluded: in state/tasks/humanoidbench.yml).
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 HumanoidBench 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✗ 58m | 09-28 19:20 | 58m | — | 18.0M / 177k | $0.317 | deterministic 1, grader_error 0, n_actions 1000, video_rendered 1 |
| L✗ 1h 00m | 10-01 15:56 | 1h 00m | — | 36.1M / 141k | $0.451 | deterministic 1, grader_error 0, live_success 0, missing_log 0, n_actions 156, replay_success 0, video_rendered 1 |
3 other run(s) of HumanoidBench
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✗ 1h 00m | 10-01 04:51 | 1h 00m | 130 | 15.1M / 107k | $3.02 | $3.13 | deterministic 1, grader_error 0, n_actions 1000, video_rendered 1 |
| L✗ 1h 00m | 09-30 23:30 | 1h 00m | 131 | 12.6M / 101k | $2.69 | $2.79 | deterministic 1, grader_error 0, live_success 0, missing_log 0, n_actions 512, replay_success 0, video_rendered 1 |
Claude Code 2.1.283 + Claude Opus 5.5, reasoning medium (OpenRouter)
Not in this run: not in this run (the two dearer models ran five tasks, chosen to span locomotion and manipulation, easy to hard)
Codex CLI 0.157.0 + GPT-6 Sol, reasoning medium (OpenRouter)closed
Not in this run: not in this run (the two dearer models ran five tasks, chosen to span locomotion and manipulation, easy to hard)
Task instruction (upstream)
Open the cabinet's compartments in order: slide open the sliding door, then pull out the drawer, then open one of the double doors and lift the cube from the drawer into that compartment, then open the flip-up door and lift the cube that started behind the double doors (not the one you put there) up into it.
What HumanoidBench states about this task
| Success criteria | 1. the summed per-step reward over one episode reaches 2500, HumanoidBench's own success bar (a total of rewards, not a number of steps; an episode is at most 1000 control steps) 2. unlimited: both fresh-process replays of the handed-in trajectory reach it and end in the same state 3. limited: the run passes the moment a live episode reaches it (the recorded episode replays to the same state); otherwise the last episode is graded |
| Env Id | h1hand-cabinet-v0 |
| Robot | Unitree H1 with two Shadow hands (61 actuators) |
| Category | Manipulation |
| Capability Class | C9 · long-horizon multi-subtask |
| Role | scored |
| Scoring | Only the compartment you are currently on is scored. Each step pays a small share for standing steadily and the bulk for progress on that one compartment; completing it pays a one-off bonus and moves the scoring on to the next. Those bonuses grow with how many you have finished, and together they are a large fraction of the bar — so stalling on the first compartment caps the episode well short of it, however cleanly that compartment is opened. |
| Ends Early | The episode ends once the last compartment is done. |
| Zero Action Return | 10.43 |
| Action Dim | 61 |
| Control Rate Hz | 50 |
| Limited Mode | head cameras (RGB 256×256), joint angles and velocities; a pelvis IMU and a camera fixed in the room when the run turns them on. Not where the robot is in the room, not the cabinet's or the cubes' state, not the reward |
| Agent Budget | 3600 s of wall clock per mode |
From https://github.com/carlosferrazza/humanoid-bench @ cb11890, as defined in our task definitions @ 7d6a7a44.
Tags¶
Why this task is interesting¶
Four compartments open in order: the sliding door, the drawer, one of the double doors (moving a cube from the drawer into it) and the flip-up door (lifting a second cube). Only the current compartment is scored, and the completion bonuses, which grow along the way, are a large part of the bar of 2500, so stalling early caps the episode well short of it.
Capability notes¶
Not yet written.
Oracle demo review¶
No demo.
Discussion¶
Stops at the first compartments: limited mode opens the sliding door, but no later compartment counts in either mode and the cubes are never touched. (@williamzhangNU)