Skip to content

Rival surface and impact-aware pace — source-only gameplay QA

Status: Implemented in the autonomous physical rival's C++ code. Portable tests and source bindings are not an Unreal Engine 5.8/UHT compile, a cooked original circuit, installed Windows gameplay, or an Android device test.

Motivation

The existing rival already observes genuine tire wear, speed, lateral slip, corner geometry and the player's moving car. However the AI's throttle/brake decisions previously did not explicitly account for the same confirmed dirt multiplier or chassis-collision damage that reduces the physical car's lateral damping. On dirt or after a severe contact, the AI could continue requesting cornering speeds calibrated for healthy pavement.

RivalSurfaceDamagePace is a conservative addition to the existing physical driving path. It does not invent terrain information, alter steering or change the official racing clock.

Production integration

The subclass ALatebrakeRivalCar::Tick reads the previous qualified chassis frame's measured ArcadeSurfaceTraction state and ArcadeCollisionDamage grip multiplier. It runs the following pipeline:

  1. Authoritative race start, real Chaos physics, valid local player and circuit gates.
  2. Approximate rival waypoint progress, recovery and legitimate corner anticipation.
  3. Tire/slip-aware RivalGripPace (existing).
  4. Confirmed off-pavement and impact-aware RivalSurfaceDamagePace (new; may only reduce throttle or apply service braking).
  5. Real player-position/velocity traffic-aware braking (existing).
  6. Existing 0.18-second following-brake hold and SetAutonomousControls into the physical base pawn.

Every source of low-grip risk is independently bounded:

  • Dirt must be previously confirmed by the existing pavement hysteresis, with its authoritative 0.72 lateral grip multiplier.
  • Impact grip must come from previously accepted chassis contacts, remain between 0.856 and 1.0, and only triggers damage-aware corner caution when it is at most 0.95 and actual steering demand is at least 0.35.
  • Invalid, stale or nonfinite samples fail closed to zero throttle and ordinary braking, with no invented acceleration.
  • On real confirmed dirt the AI begins braking when measured speed exceeds its 1,300–1,900 cm/s corner target plus 180 cm/s. On severely damaged paved corners a separately bounded 1,850–2,800 cm/s target applies.
  • Short 0.12-second braking hysteresis avoids instant brake chatter. Low-speed recovery below 650 cm/s retains normal launch throttle, so a slow or stranded rival is not permanently blocked.

These are source-level policy targets, not guaranteed real-world lap speeds: Unity-style scripted movement is not used; real Chaos forces, track contact and driveability still need installed testing.

Invariants and tests

Tests/Standalone/RivalSurfaceDamagePaceTests.cpp checks:

  • healthy paved corner stays unchanged;
  • confirmed fast dirt applies conservative braking;
  • a short hold expires, bounded throttle returns and low-speed dirt recovery is not stuck;
  • collision grip makes risky damaged turns more cautious, but a damaged straight remains free;
  • earlier route braking and checkpoint arrival remain authoritative;
  • stale road/damage states, nonfinite speed/slip, impossible grip, inconsistent dirt flags, malformed frame time and commands fail closed;
  • 200,000 deterministic valid data combinations preserve steering and never increase the rival's requested throttle.

The editor-source binding suite verifies the actual Unreal rival reads the existing surface/impact objects after tire pace and before player-traffic avoidance. Pause, restart, replay configuration, failed or successful recovery and completion clear transient braking state.

Still required before claiming this is playable

A licensed UE 5.8/UHT game build and real project-owned circuit cook; Windows gameplay with a real AI rival recovering from grass/dirt, damage and corners; player overtakes, real contact response, race reset, frame-time behavior, win/loss results and progression; hardware Android playtesting; and an evidence packet satisfying playable-release gate #148.

No source-only CI success is a substitute for these tests.