Sim benchmark · 31 models · policies · 0 submitted
Leaderboard
Full benchmark with method, sortable columns and per-model pages: bench.humanoidrobots.training
Two boards. The sim benchmark runs every published model headless each night under one generic controller, so a failure here is a fact about the model and our controller, not a verdict on the robot. The policy board ranks trained policies once they arrive with an eval we can rerun on the same pinned MJCF.
Nightly stand-hold benchmark
31/31 hold a stand · run 2026-10-11 · MuJoCo 3.14.0 WASM
| Robot | Model commit | Stand-hold 10 s | Base drift | RTF | Load | Date | Method |
|---|---|---|---|---|---|---|---|
| AgiBot X1AgiBot | e6651b9agibot_x1_train | passCoM min 98% · tilt 1.9° | -0.9 cm | 16.7× | 2.25 s | 2026-10-11 | stand-hold-v2 |
| Apptronik ApolloApptronik | c96a32dmujoco_menagerie | passCoM min 97% · tilt 1.6° | -2.3 cm | 36.7× | 4.10 s | 2026-10-11 | stand-hold-v2 |
| Berkeley Humanoid LiteUC Berkeley (Hybrid Robotics) | fc90fedberkeley-humanoid-lite-assets | passCoM min 99% · tilt 0° | -0.3 cm | 21.6× | 2.62 s | 2026-10-11 | stand-hold-v2 |
| Booster K1Booster Robotics | 12a9751booster_assets | passCoM min 99% · tilt 0° | -0.3 cm | 13.1× | 0.79 s | 2026-10-11 | stand-hold-v2 |
| Booster T1Booster Robotics | c96a32dmujoco_menagerie | passCoM min 99% · tilt 5.8° | 0.0 cm | 27.1× | 0.59 s | 2026-10-11 | stand-hold-v2 |
| Booster T1 12-DoFBooster Robotics | da396a0booster_gym | passCoM min 99% · tilt 0.3° | -0.3 cm | 45.1× | 0.59 s | 2026-10-11 | stand-hold-v2 |
| DropbearHyperspawn | dba8cd0dropbear-wbc | passCoM min 99% · tilt 2.2° | -1.3 cm | 70.7× | 0.05 s | 2026-10-11 | stand-hold-v2 |
| EngineAI PM01EngineAI | c41043fengineai_ros2_workspace | passCoM min 98% · tilt 9.6° | -0.9 cm | 18.0× | 3.87 s | 2026-10-11 | stand-hold-v2 |
| Fourier GR1-T1Fourier | 48dc081Wiki-GRx-MJCF | passCoM min 99% · tilt 0.7° | -0.4 cm | 18.4× | 0.98 s | 2026-10-11 | stand-hold-v2 |
| Fourier GR2Fourier | 11da5a7Wiki-GRx-MJCF | passCoM min 99% · tilt 0.3° | -0.3 cm | 23.2× | 0.61 s | 2026-10-11 | stand-hold-v2 |
| Fourier N1Fourier | c96a32dmujoco_menagerie | passCoM min 100% · tilt 0.6° | -0.1 cm | 19.7× | 3.20 s | 2026-10-11 | stand-hold-v2 |
| K-Scale K-BotK-Scale Labs | 08defcekbot-models | passCoM min 99% · tilt 2.2° | -0.4 cm | 25.1× | 1.09 s | 2026-10-11 | stand-hold-v2 |
| K-Scale Z-BotK-Scale Labs | f51d6eakscale-assets | passCoM min 98% · tilt 0.1° | -0.3 cm | 30.8× | 0.44 s | 2026-10-11 | stand-hold-v2 |
| Leju Kuavo 4 ProLeju Robotics | c77dc2eLeju-GMR | passCoM min 100% · tilt 1.1° | -0.2 cm | 14.6× | 1.23 s | 2026-10-11 | stand-hold-v2 |
| Menlo AsimovMenlo Research | f71f3feasimov-1 | passCoM min 99% · tilt 0.3° | -0.3 cm | 62.0× | 2.42 s | 2026-10-11 | stand-hold-v2 |
| PAL KangarooPAL Robotics | 3bc51f8kangaroo_simulation | passCoM min 94% · tilt 1.3° | -4.9 cm | 2.6× | 2.75 s | 2026-10-11 | stand-hold-v2 |
| PAL TALOSPAL Robotics | c96a32dmujoco_menagerie | passCoM min 99% · tilt 1.9° | -0.3 cm | 4.5× | 0.89 s | 2026-10-11 | stand-hold-v2 |
| PNDbotics Adam LitePNDbotics | c96a32dmujoco_menagerie | passCoM min 99% · tilt 1.5° | -0.6 cm | 25.4× | 0.21 s | 2026-10-11 | stand-hold-v2 |
| Pollen Reachy 2Pollen Robotics | f6d8284reachy2_mujoco | passCoM min 100% · tilt 0° | -0.3 cm | 10.0× | 0.93 s | 2026-10-11 | stand-hold-v2 |
| Poppy HumanoidPoppy Project | d66c8fbpoppy-humanoid | passCoM min 98% · tilt 4.3° | -0.3 cm | 18.2× | 0.08 s | 2026-10-11 | stand-hold-v2 |
| Rainbow RB-Y1Rainbow Robotics | c96a32dmujoco_menagerie | passCoM min 99% · tilt 0.1° | -0.3 cm | 25.9× | 0.72 s | 2026-10-11 | stand-hold-v2 |
| ROBOTIS OP3ROBOTIS | c96a32dmujoco_menagerie | passCoM min 99% · tilt 0.8° | -0.3 cm | 21.6× | 2.81 s | 2026-10-11 | stand-hold-v2 |
| Roboto OriginRoboparty | c71fd92roboto_origin | passCoM min 100% · tilt 0.1° | -0.3 cm | 5.2× | 2.08 s | 2026-10-11 | stand-hold-v2 |
| ToddlerBot 2XCStanford | c96a32dmujoco_menagerie | passCoM min 100% · tilt 1.6° | 0.0 cm | 9.7× | 3.43 s | 2026-10-11 | stand-hold-v2 |
| ToddlerBot 2XMStanford | c96a32dmujoco_menagerie | passCoM min 100% · tilt 1.6° | 0.0 cm | 10.1× | 2.98 s | 2026-10-11 | stand-hold-v2 |
| Unitree G1Unitree Robotics | c96a32dmujoco_menagerie | passCoM min 100% · tilt 0° | +0.2 cm | 14.5× | 1.61 s | 2026-10-11 | stand-hold-v2 |
| Unitree G1 12-DoFUnitree Robotics | 276801eunitree_rl_gym | passCoM min 99% · tilt 0.4° | -0.4 cm | 25.7× | 1.70 s | 2026-10-11 | stand-hold-v2 |
| Unitree H1Unitree Robotics | c96a32dmujoco_menagerie | passCoM min 100% · tilt 0.7° | -0.4 cm | 23.6× | 0.95 s | 2026-10-11 | stand-hold-v2 |
| Unitree H1 10-DoFUnitree Robotics | 276801eunitree_rl_gym | passCoM min 99% · tilt 0.8° | -1.1 cm | 29.9× | 1.03 s | 2026-10-11 | stand-hold-v2 |
| Unitree H1-2Unitree Robotics | 1eb6642unitree_mujoco | passCoM min 99% · tilt 0.7° | -0.4 cm | 15.7× | 1.75 s | 2026-10-11 | stand-hold-v2 |
| Unitree R1Unitree Robotics | 1eb6642unitree_mujoco | passCoM min 99% · tilt 1° | -0.3 cm | 17.1× | 1.67 s | 2026-10-11 | stand-hold-v2 |
- Method
- stand-hold-v2
- Controller
- The browser sim's hold controller (packages/sim/src/engine.ts), mirrored headless: joint servos toward the stand keyframe (torque motors run as implicit-damped servos clamped to their own torque limits, with gravity feed-forward), plus an ankle-strategy balance term from torso tilt and, where the manifest sets it, CoM-over-feet feedback; per-model gains from the @hs/sim manifest. Models published without actuators get a position servo per joint and models without a stand keyframe get one computed with the feet on the floor (each change is listed per model). No policy, no harness, no pushes.
- Pass
- Runs the full 10 s without a physics reset, the centre of mass never drops below 80% of its start height, and it ends within 10% of the start. Base drift is reported alongside.
- Fell
- The centre of mass drops below 60% of its start height at any point, or the physics diverges (NaN / auto-reset). Heights use the whole-body centre of mass because some models put the base origin near the floor.
- RTF
- Simulated seconds divided by wall seconds for the 10 s run, single thread, no rendering. Depends on the machine; see environment.
- Load
- Compile time: writing the pinned files into the WASM filesystem, mesh simplification when the model's runtime says so, MjModel.from_xml_path, and controller setup. Download time is excluded.
- Machine
- local · 13th Gen Intel(R) Core(TM) i9-13980HX ×32 · win32 10.0.26300 x64 · Node v23.11.0
- Data
- Written nightly by the Coach agent (engine) to apps/humanoidrobots.training/data/benchmarks.json and merged through a pull request.
Policy board
- Walk 0
- Walk, rough terrain 0
- Push recovery 0
- Motion imitation (dance) 0
| # | Policy | Robot | Trainer | Eval return | Sim steps | Ran on hardware |
|---|---|---|---|---|---|---|
No policies yet. Zero is the real number. Train one with mjlab and send the checkpoint, its eval video and the task id. The first entry goes here. Set up a run | ||||||