We have released RumiCar Simulator v9.0.0. The release notes below are also available on GitHub.
v9.0.0
Fixed cars getting jammed and stopping for good when many cars run together (Stage BG; this is a change to the driving physics — multi-car runs (with car-to-car contact, or where sensors see cars that finished, completed or retired) may no longer produce the same results as older versions. Runs with a single car, and runs with “other cars are obstacles” off, do not change (one exception: on precise v2, pressing ▶ again while paused no longer carries the previous run’s stuck-detection timer over). The official sample event’s verifyHash c731ecce and the frozen hashes f2 and f3 are the same as before; f0 and f1 were re-stamped). It started with a report on 2026-09-30: on the community course “サンプル8の字風 (コミュニティ例)” (a figure-8 community example), lining up 6 cars with default settings and pressing ▶ left 5 of them stopped nose to tail in the left-hand curve, never to move again. Until now the dynamic model (default) and classic resolved car-to-car contact only by “cancel that step and stop where you are”, and unlike walls there was no back-and-turn, so cars stopped in two ways: (a) they kept waiting while pushing each other in a narrow spot, and once their bodies overlapped they could never move again; (b) a car that reached the touge goal, or that completed or retired in an official race, stayed as an obstacle and the cars behind stopped behind it (measured before the change: in the live ▶ run for 120 s, 61 of the 66 shipped plus 2 community courses produced cars that stopped and never moved again at some car count). (1) Cars get out of an overlap (ESCAPE) — a car that already overlaps another car is allowed any move that does not increase the overlap area with any car (dynamic, classic). (2) Cars back and turn when a car ahead keeps blocking them (STUCK) — with “Auto-recover on crash (back & turn)” on, a car commanded forward whose move keeps being cancelled by a car ahead, and that has not moved 0.5 car lengths after 1.2 s (simulation time), starts the same back-and-turn as for a wall. It does not start one when the blocking car is behind (dynamic, classic; precise v2’s stuck detection has always done a similar job). Under “Rejoin on crash (+3s)” in official races this back-and-turn also counts as one penalty. (3) Finished, completed and retired cars are taken out — cars that reached the touge goal, and cars that completed or retired in an official race, are removed from the other cars’ sensors and collisions (all three engines). They are still drawn, so the cars behind appear to pass through them. In the live run, a car that crashed with “Auto-recover on crash” off stays where it is, as before. Effect (6 cars, measured): average stopped cars per 120 s live run on 68 courses went from 1.19 to 0.34 on lap courses and from 2.76 to 0.07 on the touge with the dynamic model, and from 1.39 to 0.18 and from 2.96 to 0.00 with classic. The finish rate in official 6-car races (3 laps, 58 races) with the dynamic model went from 28.2% to 55.7% under the retire rule and from 37.9% to 81.9% under “Rejoin on crash (+3s)”. Cars can still stop (0.34 cars per 120 s on average on lap courses with 6 cars on the dynamic model): wedged into a wall corner, unable to go forward or back, and, in official races, running the wrong way after a back-and-turn left the car facing backwards (both predate this change; the wrong-way running becomes more frequent with the back-and-turns of (2)). [Important — what no longer reproduces] Multi-car runs may no longer produce the same results as older versions (versioned records keep the version label of their time). The frozen hashes f0 and f1 (tabletop oval, dynamic model, 3 cars / 2 cars) were re-stamped to new values (the old values are kept in the history). On 3 shipped courses the check of how many cars can actually get going on a live ▶ (driveable capacity) changed (dynamic: Fictional Touge: Long Winding (Medium Slope) [2-Car Width] 3→2 cars and [4.5-Car Width] 5→6; classic: Fictional Touge: Long Winding (Gentle Slope) [2-Car Width] 2→6). How it works and what was measured are in the new §14 “Car-to-car contact and cars that stop moving” of the physics model notes, and the 📋 Specs & Limits text on contact with other cars, the “Auto-recover on crash” text, how official races count penalties and the conclusion of the steering-limit bench in §11 (3 cars at once: 1/21 → 7/21 finish) and the descriptions of the 4 bench courses were rewritten to match. The verification setup was also extended (standing verification gates 73 → 74, real-browser checks 41 → 42; the new check runs the original case — “サンプル8の字風” with 6 cars — in the production UI for 120 s and confirms that no car is stopped).
Verifying this release yourself
git clone https://github.com/RumiCar-group/RumiCar-Simulator.git
cd RumiCar-Simulator
node wf_run_all.mjs # the full assertion-gate suiteThe suite runs from a fresh clone with no dependencies beyond Node.js.
The real-browser checks under browser/ additionally need npm install and a display.