Repository navigation
Record reproducible physics replays and compare ball measurements - #6
Merged
Merged
Conversation
fred-huang122
changed the base branch from
perf/ode-0.16.6-integration
to
master
November 2, 2026 17:20
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Add a developer tool for reproducible fixed-step physics comparisons, using the same decoded protocol handlers as the live worker. It captures a fresh headless world's full settings and initial body states, seed, build/architecture/math-backend metadata, and the simulation tick at which each command or watchdog-expiry event is applied. Offline replay uses those ticks with explicit serial or helper modes, without wall-clock scheduling or rendering.
The reference comparison identifies an outstanding calibration gap:
The published numbers are from section 5.2, with a 3.3 m/s simulated shot. The comparison reports disagreement and --require-agreement exits nonzero. Its diagnostic tolerances are declared separately from experimental uncertainty; no raw uncertainty estimates are invented. Physics defaults are not fitted to another team's field. Team-specific acceleration/braking/dribbling calibration still needs measured data from that setup.
Validation: all 23 local regressions pass (86.96 s). New tests record and replay both 6 v 6 fixtures, repeat serial runs, compare helper mode numerically, detect shifted command ticks, reject malformed recordings, exercise live UDP feedback/expiry, and independently verify the measurement comparator with known errors and negative controls. A detailed helper comparison shows at most 3.4e-11 m body-position difference and 1.6e-9 rad/s angular-velocity difference over the tested pushing replay; numerical equality is not promised across CPU architectures. The pushing replay retains 0.711 mm peak shell penetration, below the unchanged 2 mm guard, with no ODE warnings/errors. The direct solver, geometry/materials, timestep and five substeps are unchanged.
Builds on the ODE changes merged in #3; replay and measurement changes are reviewed separately from rendering PR #5. Raw recordings, measurements, reports and probes remain outside Git. Only reusable tools, regressions, source-attributed reference coefficients and usage guidance are committed. Calibration against team-specific logs remains pending measured input.
A separate local experiment sets the existing rolling-deceleration option to 0.30 m/s² and observes -0.300 m/s² in the actual trajectory, matching the published rolling coefficient. Sliding acceleration and transition ratio still disagree; this does not establish a full calibration and does not change defaults.
Pre-merge review fixed three defects: Docker now stages the tools directory; unsupported safe/rolling teleport fields are rejected across the whole command batch before state mutation; explicitly requested settings must be readable regular files. Regressions cover both false/true unsupported flags with sibling robot teleports, capture/replay equality, and missing/non-file settings. The Docker build-stage source inventory now configures successfully after reproducing the original missing-source failure.
The combined reviewed rendering/replay tree passes all 27 local regressions (100.82 s) and the actual RTX 4060 application smoke check. Updated CI passes Ubuntu 22.04/24.04; native arm64 jobs are still awaiting hosted runners. The integrated tree is also being validated in PR #5 CI.