Releases / Vision + policy
Steam Powered
Shipped in openpilot 0.10.0. First public 17 Aug 2025.
Where it sits
Vision + policytwo networks
MarAprMayJunJulAugSepOctNovDecJan 2026
Tests
- G55 closed loopDistance from the recorded path at 2 snot run101520cmnot run
- G55 recovery testsAfter a 2° heading error: distance at 2 snot run405060cmnot run
- G55 open loopLateral error0.161 m/s² (0.136 to 0.169)0.150.200.25m/s²done 28 Oct 2026
Bar: this release's 95% interval, tick its mean. Grey dots: the means of the other releases in the Vision + policy family. Each axis spans every release the test has run on.
Tested 2026-10-28 04:04 UTC: protocol geval-1, harness d8eb40eeba.
Every metric, with paired differences
better, 95% interval excludes zero Better, 95% interval excludes zeroworse, 95% interval excludes zero Worse, interval excludes zerolower, 95% interval excludes zero; no better directionhigher, 95% interval excludes zero; no better direction Lower or higher, no better direction (action rate, lead)not separated, interval includes zero Not separatedno interval (too few drives) No interval: fewer than 10 drives
G55 closed loop
Not run for this model yet.
G55 recovery tests
Not run for this model yet.
G55 open loop
119 windows from 12 drives.
Measured. The G55's recorded drives replayed through the model with both of the truck's cameras (each setup's own wide-camera calibration), road-plane warp to 1.45 m. Where the narrow lens does not cover comma's narrow view (the 8 mm lens in the V13 case), those pixels come from the wide camera, brightness matched. Truth is the truck's CAN speed and yaw rate. 30 s windows after a 5 s warm-up; window = unit, drive = cluster; results are also given per camera setup. Corpus windows of 30 s from the truck's recorded drives, scored per camera setup.
| Metric | Mean | vs lebowski | vs Filet o Fish |
|---|---|---|---|
| Lateral errorm/s², lower is better | 0.161[0.136, 0.169]median 0.155 | not separated, interval includes zero−0.012[−0.024, +0.009]better on 72/119 | better, 95% interval excludes zero−0.037[−0.044, −0.020]better on 110/119 |
| Lateral error at the model's own best leadm/s², lower is better | 0.156[0.133, 0.167]median 0.153 | not separated, interval includes zero−0.008[−0.020, +0.013]better on 70/119 | better, 95% interval excludes zero−0.037[−0.044, −0.019]better on 108/119 |
| Lateral correlation with the car's later motionhigher is better | 0.837[0.805, 0.855]median 0.920 | better, 95% interval excludes zero+0.070[+0.030, +0.085]better on 101/119 | better, 95% interval excludes zero+0.017[+0.008, +0.032]better on 98/119 |
| Lateral gain (1 = same size as the car's motion)closer to 1 is better; the delta is the change in distance from 1 | 0.874[0.839, 0.940]median 0.887 | not separated, interval includes zero−0.055[−0.082, +0.007]better on 78/119 | better, 95% interval excludes zero−0.022[−0.058, −0.007]better on 71/119 |
| Lateral action rate, RMS (smoothness)m/s³, no better direction | 0.652[0.386, 0.724]median 0.537 | lower, 95% interval excludes zero; no better direction−1.638[−1.895, −0.813]lower on 119/119 | lower, 95% interval excludes zero; no better direction−0.932[−1.074, −0.651]lower on 119/119 |
| Lead of the lateral request over the car's motions, no better direction | 0.693[0.635, 0.724]median 0.650n = 118 | not separated, interval includes zero+0.075[−0.024, +0.173]lower on 27/118 | higher, 95% interval excludes zero; no better direction+0.108[+0.041, +0.143]lower on 14/116 |
| Longitudinal errorm/s², lower is better | 0.346[0.286, 0.389]median 0.329 | not separated, interval includes zero−0.023[−0.041, +0.031]better on 74/119 | not paired |
| Longitudinal correlationhigher is better | 0.335[0.236, 0.538]median 0.294 | not separated, interval includes zero+0.008[−0.111, +0.055]better on 60/119 | not paired |
| Longitudinal error while brakingm/s², lower is better | 0.732[0.532, 0.815]median 0.718n = 81 | not separated, interval includes zero+0.234[−0.003, +0.334]better on 17/81 | not paired |
| Longitudinal action rate, RMS (smoothness)m/s³, no better direction | 0.570[0.456, 0.626]median 0.537 | lower, 95% interval excludes zero; no better direction−5.039[−6.093, −2.762]lower on 119/119 | not paired |
| Plan speed error at 3 sm/s, lower is better | 2.72[1.51, 3.42]median 2.69 | worse, 95% interval excludes zero+0.72[+0.27, +1.42]better on 25/119 | not separated, interval includes zero+0.29[−0.13, +0.45]better on 35/119 |
By G55 camera setup (5 setups)
| Metric | V14 (8 mm + IMX662)8 win, 2 drv | V13 (8 mm + IMX662)92 win, 3 drv | Stock aimed up4 win, 2 drv | Stock new case2 win, 1 drv | Stock dual13 win, 4 drv |
|---|---|---|---|---|---|
| Lateral errorm/s² | 0.150med 0.152 | 0.167med 0.161 | 0.108med 0.108 | 0.166med 0.166 | 0.133med 0.127 |
| Lateral error at the model's own best leadm/s² | 0.148med 0.147 | 0.163med 0.160 | 0.104med 0.104 | 0.162med 0.162 | 0.130med 0.126 |
| Lateral correlation with the car's later motion | 0.856med 0.918 | 0.833med 0.927 | 0.837med 0.928 | 0.846med 0.846 | 0.854med 0.865 |
| Lateral gain (1 = same size as the car's motion) | 0.905med 0.908 | 0.868med 0.883 | 0.852med 0.987 | 0.805med 0.805 | 0.914med 0.893 |
| Lateral action rate, RMS (smoothness)m/s³ | 0.437med 0.381 | 0.732med 0.671 | 0.369med 0.377 | 0.527med 0.527 | 0.329med 0.274 |
| Lead of the lateral request over the car's motions | 0.681med 0.675 | 0.698med 0.650 | 0.750med 0.775 | 0.525med 0.525 | 0.673med 0.650 |
| Longitudinal errorm/s² | 0.361med 0.325 | 0.338med 0.328 | 0.618med 0.606 | 0.452med 0.452 | 0.288med 0.253 |
| Longitudinal correlation | 0.727med 0.716 | 0.275med 0.242 | 0.307med 0.330 | 0.495med 0.495 | 0.505med 0.625 |
| Longitudinal error while brakingm/s² | 0.697med 0.733 | 0.745med 0.731 | 1.078med 1.078 | 0.223med 0.223 | 0.691med 0.680 |
| Longitudinal action rate, RMS (smoothness)m/s³ | 0.523med 0.482 | 0.600med 0.553 | 0.543med 0.467 | 0.579med 0.579 | 0.399med 0.372 |
| Plan speed error at 3 sm/s | 1.36med 1.40 | 3.10med 3.04 | 2.00med 2.23 | 2.51med 2.51 | 1.10med 1.06 |
Limits
- Open-loop scores compare a model's action with what the recording driver or car did; a model that drives differently but well is scored as wrong. comma's RL residual makes the lateral action smoother, and earlier than its base's by 0.10 to 0.26 s (measured on the G55 and on 16 of comma's cars); both score as error here. Measured on the 2026-10-27 deep dive: open-loop accuracy did not predict closed-loop driving.
- Closed-loop numbers come from a generative world model, not a car. In a separate test on the G55's windows, worldmodel-4B followed an added yaw command through 2 s but showed 0 to 20% of an added lateral offset 3 to 4 s later; past about 4 s the recorded anchor frames also pull every model back toward the recorded path. The front page reports 2 s.
- Intervals are 95% percentile bootstraps that resample whole drives (the drive a public segment was cut from, the recorded drive a truck window comes from), so segments of one drive count once. Below 10 drives there is no interval: the braking-lead test comes from one drive and reports means and win counts only.
- Paired deltas are model minus baseline on the units both have (for lateral gain, the change in its distance from 1); wins count units where the model is better in the metric's direction.
- The lead of the lateral request is the shift, searched from 0 to 1.5 s, at which the request best correlates with the car's later lateral motion; a larger value means the request comes earlier. A value at either edge is censored and left out. Plan speed error at 3 s reads the plan head, which does not separate models built on one frozen base: it is identical on every segment for the four f22e06e5 models, and for Cinque v2 and v3.
- Truck open loop comes from few drives, and the camera setup changed between them (lens, housing, aim); the pooled number mixes setups, so read the per-setup numbers too. The 8 mm setups' wide camera recorded about 7-10 fps, so its frame can be up to ~60 ms from the narrow one. On the V14 setup up to 7 of a window's 700 steps are ~100 ms (one camera frame dropped). The camera's aim is the truck's live calibration learned on each drive; drives whose recorded calibration was carried over from another drive are left out.
- A historical model, run outside comma's software with its own release's inputs, camera warp and command (engines/legacy.py). Checked against comma's CI model replay of this model (made on a comma device, 60 frames of one route): the plan's lateral position is 0.040 m RMS from comma's output; every other model here is 0.83 m or more from it on the same check.
- Trained for the comma devices and cameras of its time, not for the G55's camera or for every camera among comma's public segments: of those 45 segments 26 record the OX03C10 sensor, 6 the AR0231, 4 the OS04C10 and 9 none; 30 were recorded 2021 to 2024 and 15 come from devices whose route names carry no date.
- Not run in closed loop: the G55 closed-loop and recovery tests need this release's controller inside the world-model loop, which is not built yet.
- The commands are 0.10's modeld (get_action_from_model on the plan, lateral delay = steer actuator delay + 0.2 s, smoothed over 0.1 s lateral and 0.3 s longitudinal); the newest models are scored on their own unsmoothed action head, so lateral action rates are not comparable across these eras.
The model file
- Model id
steam_powered- SHA-256
e66bb8d53eced3786ed71a59b55ffc6810944cb217f0518621cc76303260a1ef+04b763fb71efe57a8a4c4168a8043ecd58939015026ded0dc755ded6905ac251- File tested
- openpilot-master,
ceb557058c5a(source), 2025-08-17 - Parameters
- not counted
- GMAC per step
- n/a
- Notes
- Shipped in openpilot 0.10.0. Plan only: modeld turns the plan into curvature and acceleration (get_action_from_model), so both lateral and longitudinal are scored. Vision 2454c1c0/500, policy a8f96b93/400.
Results file (results/steam_powered.json)
History of this page
- Tests run (1 of 3), results published
- Steam Powered merged to openpilot master