Guide

Game vehicle physics explained

Harbor Arena's Season 3 coastal circuit looked spectacular in trailers — until the first multiplayer stress test. Cars launched off the cliff chicane at 240 km/h, landed with four wheels in the air, and snapped sideways when the suspension finally caught pavement. The racing design called for forgiving drift corners; the underlying vehicle physics was tuned like a stiff sim racer on a 120 Hz fixed timestep with no downforce and wheel colliders that teleported through curbs. The refactor replaced naive force-at-center steering with per-wheel slip curves, retuned spring-damper rates for curb strikes, added a short airborne stability assist, and layered an arcade handbrake yaw boost only above a slip threshold. Lap-time variance between skilled and casual players narrowed; rollover complaints dropped to near zero. Vehicle physics is the stack that turns throttle and steer input into believable motion: a rigid chassis, spring-suspended wheels, engine torque delivered through a gear model, lateral grip that falls off with slip angle, and optional assists that trade realism for fun. This guide covers the arcade-vs-simulation spectrum, wheel collider fundamentals, suspension and contact patches, powertrain tuning, steering and drift behavior, aero and stability, engine-specific patterns in Unity and Unreal, a Harbor Arena coastal circuit walkthrough, an approach decision table, common pitfalls, and a production checklist.

Arcade vs simulation: pick a lane first

Every vehicle system sits on a spectrum. At the arcade end, the car is a responsive toy: high baseline grip, strong auto-countersteer, capped angular velocity, and maybe a boost button. Players feel heroic within minutes. At the simulation end, weight transfer, tire temperature, brake bias, and aero balance matter; mastery takes hours but rewards precision. Most shipped games are sim-lite: real wheel contacts and suspension, but exaggerated assists and forgiving slip windows.

Decide your target before tuning numbers. A battle-royale buggy and a time-trial GT car can share an engine module but need different assist profiles. Document which assists are intentional (handbrake pivot, traction control, steering sensitivity curves) so playtest feedback does not get misread as "physics bugs."

Chassis rigid body and center of mass

The vehicle body is typically a single rigid body with collider geometry simplified to a box or convex hull. Mass, center-of-mass (CoM) height, and inertia tensor dominate handling feel more than mesh fidelity. A CoM too high produces rollovers on fast direction changes; too low and the car feels glued and refuses to pitch under braking.

Lower the CoM slightly below the visual cabin for arcade racers; raise it carefully for off-road trucks that should lean. Use physics materials for chassis-ground friction only when the body can scrape — wheel contacts should carry most longitudinal and lateral forces. Interpolation on the rigid body smooths render frames when physics ticks at 50–60 Hz while rendering at 120 Hz.

Wheel colliders and suspension

Production engines expose wheel colliders (Unity WheelCollider, Unreal Chaos Vehicle wheels) that raycast or sweep downward each substep, apply spring-damper forces along the suspension axis, and compute longitudinal/lateral friction at the contact patch.

Spring-damper tuning

Suspension is a spring (stiffness) plus damper (compression/rebound). Too stiff: wheels bounce off curbs and lose contact. Too soft: the body wallows, delayed weight transfer, and mushy turn-in. Target roughly 1:10 damper-to-spring ratio as a starting point, then iterate on curb strikes and landing events. Suspension distance sets travel; force application point should be at the wheel hub, not the body center.

Wheel radius and visual sync

Mismatch between collider radius and rendered wheel mesh causes foot sliding and incorrect ride height. Each frame, rotate the visual wheel mesh by rpm × deltaTime and steer the front pair by the steer angle returned from the physics API. Grounded checks gate engine torque and sound — spinning in air should not apply drive force.

Powertrain: torque, gearing, and braking

Engine output is a torque curve over RPM, not a flat horsepower number. Arcade games often fake this with piecewise curves: strong low-end for launch, taper at top speed. Multiply engine torque by gear ratio and final drive, then divide by wheel radius to get drive force at the contact.

Gear shifts can be automatic (shift up at RPM threshold, downshift on throttle + low RPM) or player-controlled. Add shift delay and brief torque cut to avoid instant upshift exploits. Braking applies negative longitudinal force per wheel; front-biased brake balance helps straight-line stops, rear bias enables rotation under trail braking in sim-lite titles. Handbrake usually locks or reduces rear lateral grip — a design lever for drift, not a literal parking brake simulation.

Steering, grip, and slip angles

Lateral force at each tire depends on slip angle — the difference between where the wheel points and where it actually travels. A simplified Pacejka-style curve rises with slip up to a peak, then falls (breakaway). Arcade tuners widen the peak plateau so casual players stay in the grip region; sim titles narrow it and punish overdriving.

Steering input should pass through a sensitivity curve (slow speeds: more angle per stick unit for tight parking; high speeds: reduced max steer to prevent snappy spins). Ackermann geometry on front wheels helps low-speed turning radius; at speed, many games reduce inner-wheel angle artificially. Pair steering with speed-scaled yaw assist only if your design doc allows it — invisible assists frustrate sim players.

Drift and handbrake assists

Drift-friendly games temporarily raise rear slip limits or add yaw torque when handbrake is held above a speed threshold. Cap assist duration so players cannot infinite-slide every corner. Surface types (asphalt, wet, dirt) multiply peak grip — a single global friction value rarely survives art-directed biomes.

Aero, downforce, and airborne stability

At high speed, apply downward force proportional to velocity squared to increase grip without changing low-speed handling. Front/rear aero split tunes understeer vs oversteer balance on straights. Keep coefficients modest unless you are building a dedicated sim — excessive downforce makes slow-speed hairpins feel numb.

When all wheels lose contact, apply mild angular damping or align torque toward velocity to prevent tumbling after jumps. This is an assist, not realism — disclose it in design docs. Landing detection can briefly stiffen dampers to absorb impact without launching the chassis.

Engine patterns: Unity, Unreal, and custom raycast

Unity: WheelCollider on child objects, motor torque via motorTorque, steer via steerAngle, read rpm, isGrounded, and GetGroundHit for surface response. Use a separate input layer mapping gamepad triggers through input curves.

Unreal Engine: Chaos Vehicle or legacy PhysX vehicle component with engine setup assets defining torque curves, transmission, and differential (open, locked, LSD). Chaos improves determinism and multi-body coupling but still needs per-project grip tuning.

Custom raycast vehicles cast four rays from hubs, apply forces manually, and run on deterministic logic — common in mobile and networked racers where engine wheel colliders are opaque or non-deterministic. More code, full control, higher maintenance.

Harbor Arena coastal circuit refactor (worked example)

Problem: cliff chicane launches and curb snagging caused rollovers and random spin-outs in Harbor Arena's Season 3 coastal track.

CoM and collider: Lowered CoM 12 cm, replaced tall mesh collider with low box hull; body scrape friction reduced so curbs push rather than trip.
Suspension: Increased travel 8 cm, softened spring 15%, raised rebound damping on rear axle to control landing pitch.
Per-surface grip: Asphalt, wet sand, and curb tags multiplied peak lateral force; wet sand added mild understeer bias.
Slip curves: Widened arcade plateau to 8° slip; breakaway softened so recoveries possible without full spin.
Airborne assist: 0.4 s angular damping when all wheels airborne; downforce ramp starting at 100 km/h.
Handbrake drift: Yaw boost only if speed > 60 km/h and rear slip > threshold; disabled in first-lap tutorial.

Result: median lap time unchanged for top quartile; bottom quartile completed races without rollover DNF; telemetry showed 62% fewer full 360° spins on the chicane.

Approach decision table

Goal	Recommended approach	Trade-off
Casual mobile racer	High grip, auto steer assist, simple raycast wheels	Low skill ceiling; sim players bounce off
Console arcade racer	Engine wheel colliders + handbrake drift + downforce	Tuning time on curbs and jumps
Sim-lite PC title	Full torque/gears, surface types, optional assists off	Controller and wheel support expected
Networked multiplayer	Deterministic custom raycast or fixed-step Chaos	Engine defaults may desync across clients
Open-world driving	LOD physics (simple at distance), streaming terrain collision	Handoff bugs at streaming boundaries
Off-road / trucks	Long suspension travel, diff lock, low-speed torque	High CoM rollover risk without assists

Common pitfalls

Applying drive torque when airborne. Gate motor force on grounded wheel count; otherwise cars accelerate mid-jump.
Steering the rigid body directly. Yaw torque on the chassis fights wheel friction and feels floaty; steer the front wheels.
Identical front/rear grip. Produces bland neutrality; slight rear bias enables fun rotation in arcade titles.
Ignoring fixed timestep. Variable deltaTime on vehicle forces causes frame-rate-dependent handling.
Visual wheel desync. Players trust what they see; sliding tires without smoke/audio reads as broken physics.
Curbs as sharp vertical meshes. Triangle edges catch wheel rays; bevel curbs or use simplified collision.
One tuning profile for all vehicles. Weight classes need distinct spring rates and torque peaks.
Network replication of every substep. Replicate input and low-rate state; simulate locally where possible.

Production checklist

Document arcade vs sim-lite target and which assists are enabled.
Set CoM, mass, and inertia per vehicle class before grip tuning.
Match wheel collider radius and suspension rest length to mesh.
Tune spring/damper on curb strikes, landings, and flat cornering.
Plot torque curve and verify top speed matches design on longest straight.
Calibrate steer sensitivity at 30, 90, and 180 km/h reference speeds.
Define surface grip multipliers for each biome material tag.
Test handbrake/drift assist on controller and keyboard separately.
Record telemetry: slip angle, per-wheel load, air time, lap variance.
Run multiplayer desync test on jump-heavy and curb-heavy sections.

Key takeaways

Vehicle feel comes from CoM, suspension, per-wheel grip, and powertrain curves — not mesh detail.
Pick arcade or sim-lite upfront; assists are design tools, not failures.
Wheel colliders handle contact; steer wheels, do not yaw-torque the chassis as primary input.
Slip-angle grip curves and surface types separate predictable handling from random spin-outs.
Validate curbs, jumps, and network determinism before polishing VFX.