Skip to content

Manipulation · Truck

keephumanoidbench_manip——@williamzhangNU

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

TrialStartedAgent timeModel requestsTokens in / outEst. costGrade
U✗ 59m10-04 06:2859m—19.6M / 110k$0.266deterministic 1, grader_error 0, n_actions 1000, video_rendered 1
L✗ 7m10-04 06:307m—2.2M / 38k$0.047deterministic 1, grader_error 0, live_success 0, missing_log 0, n_actions 1000, replay_success 0, video_rendered 1
3 other run(s) of HumanoidBench

Codex CLI 0.159.2 + GPT-6.1 Sol, reasoning medium (OpenRouter)

TrialStartedAgent timeModel requestsTokens in / outEst. costBilledGrade
U✗ 1h 00m10-01 04:481h 00m12013.3M / 108k$2.80$2.91deterministic 1, grader_error 0, n_actions 1000, video_rendered 1
L✗ 50m10-01 00:2750m10210.4M / 87k$2.87$3.12deterministic 1, grader_error 0, live_success 0, missing_log 0, n_actions 1000, 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)

Unload the five loose packages from the truck (the wall of boxes behind them is fixed) and put them on the table — all of them.

humanoidbench_manip scene
No oracle video published upstream — this is the scene it starts from.
What HumanoidBench states about this task
Success criteria 1. the summed per-step reward over one episode reaches 3000, 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 its one episode reaches it (no reset in our runs since 2026-10-04; the recorded episode replays to the same state)
Env Id h1-truck-v0
Robot Unitree H1 (19 actuators)
Category Manipulation
Capability Class C9 · long-horizon multi-subtask
Role scored
Scoring Each step is your uprightness multiplied by a base amount plus three progress terms: how close the robot is to a package still on the truck, how close it is to a package it has picked up, and how close a picked-up package is to the table. On top of that, lifting a package off the truck and placing one on the table each pay a one-off bonus (taken back if a package comes off the table again), and delivering the last package pays a larger one and ends the episode. Uprightness is a factor, so lying down costs a large part of every step. The one-off bonuses are most of the bar but not all of it, so the rest has to come from the per-step terms before the last delivery.
Zero Action Return 544.42
Action Dim 19
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 where the packages are, 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 @ 841996303.

Tags

Why this task is interesting

Unload the five loose packages from a truck about 3 m away, up a ramp, and put them on a table. Lifting a package and placing it pay one-off bonuses on top of the per-step terms, so standing well earns little of the bar.

Capability notes

Not yet written.

Oracle demo review

No demo.

Discussion

Blocked on walking: the robot never gets up the ramp to the packages; GPT-6 Luna (2026-10-04) stood in unlimited mode, and in limited mode it fell and the episode ran on to its 1000 steps (truck ends only when the last package is on the table). (@williamzhangNU)