Skip to content

Rival yaw-rate steering moderation — source-only gameplay QA

Scope: The actual physical Unreal racing rival now uses its measured Chaos chassis angular velocity to moderate over-rotation in high-speed turns. This source change is not a licensed UE 5.8/UHT compile, a verified racing game, a cooked original circuit, a Windows EXE or an Android APK.

Motivation

The existing PlanRivalTurn computes a desired heading and steering input from the next physical circuit waypoint. The base car applies real chassis yaw torque. While cornering quickly, the rival may already be rotating in the desired direction, yet keep requesting nearly the same steering torque until its heading catches up. That can contribute to overshoot and unnecessary correction.

ModerateRivalYawRate provides one conservative feedback step: reduce desired steering only when the chassis's real world-Z angular velocity has the same sign as requested steering, and both yaw rate and planar speed are elevated. It does not add opposite steering, steer around other cars, or change waypoint selection.

Actual C++ pawn integration

The qualified ALatebrakeRivalCar::Tick runs:

  1. Authoritative race/start/physics/circuit gates and real chassis position/speed observation.
  2. Waypoint steering, corner anticipation, tire grip adaptation, and confirmed off-pavement/impact pace adjustment.
  3. Real player-traffic braking and short following-brake hold.
  4. Real Chassis->GetPhysicsAngularVelocityInRadians().Z sampling and ModerateRivalYawRate application to the final pilot command.
  5. Existing SetAutonomousControls, which routes through the normal player/rival base pawn force/steering and speed budgets.

Conservative tuning

  • At or below 950 cm/s actual planar speed, steering remains unchanged.
  • At or below 0.30 rad/s yaw, steering remains unchanged.
  • Between 950–3150 cm/s and 0.30–1.50 rad/s, attenuation rises continuously as both measurements increase.
  • Maximum attenuation is 65%, retaining at least 35% of the previously requested steering magnitude for high-speed correction.
  • Opposite-direction measured yaw retains 100% of requested corrective steering. Neutral input never creates countersteer.
  • Existing brake decisions, throttle, waypoint arrival, and real player-traffic policy are copied unchanged.
  • Invalid physics readiness, malformed/nonfinite measurements, frame hitches >0.10 s or invalid pilot commands fail closed to neutral steering, zero throttle and a brake request.

This is not a Chaos vehicle steering actuator, physical stability control, real suspension, a per-wheel tire model or a promise of better real-world lap times.

Portable regression and source bindings

The standalone RivalYawRateSteeringTests.cpp checks:

  • Same-direction left/right yaw moderation, symmetry and sign retention.
  • Full steering at launches, ordinary yaw, and when correcting opposite-direction rotation.
  • No generated steering from a neutral command; no changes to braking, throttle, or arrived-waypoint decisions.
  • Finite input ranges, invalid chassis samples, and failed readiness.
  • Monotone yaw attenuation and 200,000 deterministic motion samples preserving all existing pilot decisions except steering magnitude.

The editor-source bindings ensure the real rival takes the yaw sample from its physical Chaos chassis after corner and traffic policies and before the existing SetAutonomousControls call.

Real engine/hardware acceptance still required

  1. Build UE 5.8 Editor/UHT and the game target with valid owned assets, and author/cook the original LatebrakeCircuit map.
  2. Drive an actual rival through left/right hairpins and straights at low and high speed. Compare real chassis yaw/angular velocity, overshoot and lap progress against the earlier build.
  3. Verify a recovering or spinning rival retains enough directional steering to get back on route; ensure no oscillation or indefinite braking with tire, dirt, and collision damage systems.
  4. Test traffic following, collisions, focus loss, pause, startup, restart, recovery teleport, and malformed-physics fail-closed paths.
  5. Capture installed Windows and Android traces and qualification evidence under playable-release gate #148.

Portable source tests alone do not certify real installed driving.