Skip to content

Rival grip-aware corner pace — source-only QA

Source-only prototype. Not a verified Unreal Engine 5.8 game build. The existing AI follows the same directional circuit waypoints; this change does not create an overtaking system, a separate racing line, actual per-wheel tire dynamics, authoritative opponent standings or a pit strategy.

Gameplay behavior

  • The autonomous rival already receives the exact same race-local chassis-derived arcade tire heat and wear as the player. The new RivalGripPace uses its previous valid physics-tick grip, current real chassis planar/lateral velocities, steering demand and delta time.
  • After existing route steering/corner anticipation and before existing player-aware traffic braking, a grip-risk governor can request ordinary braking or decrease throttle. The policy may never increase drive, alter steering, bypass a waypoint or weaken a brake.
  • A turn requires speed at least 11 m/s, steering magnitude at least 0.22, and either 3%+ loss of available arcade grip or measured lateral slip of at least 12%. Stronger slip, tighter steering and worn tires lower a conservatively bounded target pace. If actually overspeed by more than 1.5 m/s, ordinary braking takes priority.
  • A 0.12-second maximum brake hold dampens brake/release chatter while a cornering hazard subsides. It resets on race restart, pilot reconfiguration, route completion, recovery and blocked/paused physics.
  • Malformed or stale velocities, grip, control commands or frame deltas fail closed (no motor demand, safe braking) and clear transient state. Missing tire telemetry also prevents the AI from relying on invented grip samples.
  • The existing player-aware rival following, event countdown, four-gate route recovery, damage, lap clock, credits, car garage, nitro, drafting and physical chassis collision remain separate authorities.

Portable and source-binding verification

cmake -S Tests/Standalone -B Tests/Standalone/build -DCMAKE_BUILD_TYPE=Release
cmake --build Tests/Standalone/build --config Release --parallel
ctest --test-dir Tests/Standalone/build -C Release -R '^RivalGripPace$' --no-tests=error --output-on-failure
python3 -m unittest discover -s Tests/Editor -p 'RivalGripPaceBindingTests.py' -v

The portable tests cover valid grip-limited straight/corner/launch cases, brake hysteresis, already-braking and arrived routes, time hitches, nonfinite telemetry, invalid controls and 5,000 race-session reset cycles. Python source tests enforce ordered integration with the original corner anticipation, live tire condition, traffic awareness and rival control path. These are not Chaos physics simulations.

Mandatory installed-game acceptance

  1. In licensed Unreal Engine 5.8, compile actual UHT and Game/Editor code; cook the authentic owned race map and package a real candidate.
  2. Drive a complete multi-lap race on Win64. Observe the actual AI at straight entries, high-speed corners, cold and worn race-session tires, right/left approaches, player traffic and final waypoint.
  3. Confirm low grip causes only slower corner-entry pace, never arbitrary steering or acceleration. Compare at 30, 60 and 120 FPS; check for stuck/braking loops and realistic pit-free completion.
  4. Check enemy route/recovery, interrupted focus, pause, failed physics, stopped player, restart and race-end handling. The AI must not acquire additional official wins, rewards or save privileges.
  5. Record exact build SHA, engine version, map, package hash, logs, controller/keyboard evidence and failures before claiming gameplay.

This is source-only until UE compile, map cook and installed-device testing are documented on playable gate #148.

Cross-policy composition regression (next source development cycle)

The additional RivalRacecraftComposition portable CTest checks the same source decision order as the physical rival's Tick implementation: route/corner planning → actual observed grip/sideways velocity → player-lane traffic awareness → short traffic-brake hold → motor command.

The test covers independent corner and traffic braking, eventual brake-hold expiry, checkpoint arrival authority, missing tire/player measurements, nonfinite frame time and 100,000 bounded, consecutive numeric frames. Its Python source binding companion checks that the Unreal AI pawn still calls the policies in the expected order and preserves existing steering.

The fixture does not render or run a vehicle, calculate actual Chaos contact, qualify road surface grip, prove the automated racer can overtake or verify checkpoint/collision gameplay. A licensed UE 5.8 build and installed Windows/Android driving assessment are required before a playable, stable, or competition-ready claim.