Skip to content

Velocity-consistency candidate v11 results: Go (PPC 0/576, UrbanNav 0/192)

The frozen contract is online_pva_candidate_v11.md.

  • Freeze: commit 6f2fab94, before any comparison replay.
  • Contract SHA256: f685b873c2d34e4f0d77c7962063cea1ff63b61511c3f620fee6c06104eb140c.
  • Default: unchanged. Go permits a separate, reviewed default-switch proposal.
  • Tuning: nothing was tuned after the results were seen.
  • Machine-readable record: online_pva_decision_v11.json.

Decision

Go. Every gate passes on all 24 run/scenarios.

  • PPC: 0 of 576 gates fail. That is the 558 earlier gates plus gate 8 on 18 run/scenarios.
  • UrbanNav: 0 of 192 gates fail.
  • Targeted rotation improvement: present in both comparator outputs.
  • Gate 8 (attitude integrity, absolute): the candidate has no epoch with rotation error above 90 deg in any run/scenario. The largest rotation error anywhere is 38.5 deg (Shinjuku IMU gap). The control fraction above 90 deg is 0-55 %, depending on the run/scenario.
  • Gate 7: control none is bit-identical to develop a7a5d5bf on the 18 PPC run/scenarios: 175,902 rows, every deterministic CSV field.

Run details:

  • Replays: 48 replays from one binary (SHA256 854bdd42..., built from the frozen worktree), interleaved control/candidate, at most 3 at once. All passed and none were rerun.
  • Processor P95, candidate/control: min 1.11, mean 1.36, max 1.92. The limit is 2.
  • UrbanNav inputs are read with the RINEX 3.00-3.02 BeiDou B1I fix (#577). UrbanNav control values therefore differ from the records of v10 and earlier.

Normal runs, control -> candidate

Values are RMSE / P95, except fused velocity, which is RMSE.

Run RTK pos m RTK vel m/s Fused pos m Fused vel m/s Rotation deg
Tokyo 1 32.2/61.7 -> 21.1/34.7 2.10/3.29 -> 1.17/2.22 71.7/155.2 -> 4.0/6.2 7.05 -> 0.18 105.1/171.1 -> 1.7/2.4
Tokyo 2 19.5/29.7 -> 6.3/10.7 2.26/3.71 -> 0.75/1.59 74.2/179.5 -> 2.4/5.1 6.85 -> 0.10 100.3/170.0 -> 1.4/1.9
Tokyo 3 18.6/31.9 -> 12.0/24.0 2.10/1.76 -> 0.66/1.30 41.9/115.0 -> 10.4/9.0 3.22 -> 0.15 63.5/144.4 -> 1.6/2.6
Nagoya 1 25.7/29.0 -> 20.9/26.7 2.51/4.26 -> 1.28/2.36 49.3/118.2 -> 12.5/11.3 4.61 -> 0.88 88.2/168.7 -> 2.9/3.0
Nagoya 2 37.7/81.8 -> 30.9/71.7 2.30/5.09 -> 1.49/2.82 81.0/168.0 -> 7.3/17.1 3.73 -> 0.33 18.5/39.8 -> 2.1/4.3
Nagoya 3 43.7/101.1 -> 37.0/99.5 2.47/5.13 -> 1.50/3.31 58.9/128.1 -> 13.6/19.9 5.54 -> 0.20 105.0/170.9 -> 1.8/2.8
Odaiba 199.4/27.7 -> 199.0/26.7 3.06/7.17 -> 0.38/0.51 82.7/211.7 -> 11.2/12.6 6.96 -> 0.21 105.7/171.8 -> 2.1/3.3
Shinjuku 27.0/64.9 -> 16.6/31.7 2.17/4.99 -> 0.76/0.75 136.3/323.8 -> 7.0/11.2 24.41 -> 0.28 99.7/170.5 -> 2.2/3.7

Against velocity_consistency_v9 (rotation, normal scenario)

  • PPC (RINEX 3.04, comparable):
  • Tokyo 1-3 and Nagoya 2-3 are equal or better.
  • Nagoya 1 is worse, 1.37 -> 2.90 deg RMSE. The contract predicted this. The cost sits in the first ~100 s, and the run stays far below the control's 88 deg.
  • UrbanNav: the earlier v9 records were made before the reader fix, so they are not comparable. On the same fixed input, v9 lost the attitude on Odaiba and Shinjuku (about 105 deg RMSE). v10 holds it at 2.1 and 2.2 deg.

Remaining limits

  • Development data only. All eight runs are development data, and the diagnosis ran full counterfactuals on them.
  • Odaiba RTK position RMSE is about 199 m in both arms. The cause is a few very large RTK outliers, which this candidate does not address.
  • A default switch needs its own reviewed proposal. The proposal should include one run on data never used in development. The Hong Kong conversion prepared for holdout v2 has not been run with any estimator.