Locomotion · Hurdle¶
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✗ 59m | 10-04 05:24 | 59m | — | 51.5M / 268k | $0.738 | deterministic 1, grader_error 0, n_actions 1000, video_rendered 1 |
| L✗ 1m | 10-04 05:24 | 1m | — | 246k / 6k | $0.0080 | deterministic 1, grader_error 0, live_success 0, missing_log 0, n_actions 49, 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✗ 59m | 10-01 02:39 | 59m | 107 | 8.7M / 96k | $2.15 | $2.23 | deterministic 1, grader_error 0, n_actions 1000, video_rendered 1 |
| L✗ 30m | 09-30 22:25 | 30m | 73 | 5.5M / 62k | $1.45 | $1.52 | deterministic 1, grader_error 0, live_success 0, missing_log 0, n_actions 105, 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)
Run the robot forward down the walled track and over the hurdles, without touching the side or back walls and without falling.
What HumanoidBench states about this task
| Success criteria | 1. the summed per-step reward over one episode reaches 700, 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-hurdle-v0 |
| Robot | Unitree H1 (19 actuators) |
| Category | Locomotion |
| Capability Class | C2 · constrained gait |
| Role | scored |
| Scoring | Forward speed, staying upright (head at standing height, torso vertical) and low actuator force, multiplied together — but the speed target is a sprint rather than a walk, on the order of 5 m/s. On top of that, touching any barrier's collision geometry multiplies that entire step down to a small fraction of its value: contact is a gate to be respected, not a penalty to be traded off. Only the walls that line the track on both sides and behind the start count as barriers here; the hurdles themselves carry no such penalty, but they are solid and have to be cleared. |
| Ends Early | The episode ends early if the pelvis drops near the ground. |
| Zero Action Return | 17.9 |
| 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 on the track, not where the hurdles and walls 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¶
A sprint of about 5 m/s down a walled track, with hurdles from x = 7 m. The robot has to run fast and stably first, then clear each hurdle without touching the walls or falling.
Capability notes¶
Not yet written.
Oracle demo review¶
No demo.
Discussion¶
Speed is the gap: GPT-6.1 Sol's unlimited run covers the whole episode without falling or touching a wall but at about 1.4 m/s, and GPT-6 Luna's (2026-10-04) stops short of the first hurdle; limited mode never gets past 1.92 m, and without resets Luna's episode ended in a fall at step 49. (@williamzhangNU)