Aetherfall: High-Performance Flight Physics & Atmospheric Engine
Delivered 120 FPS deterministic aerodynamics and custom volumetric cloud shading on PC & console.
Aetherfall fighter aircraft cruising through procedural volumetric cloudscape
01 / The Challenge & Constraints
Modern flight simulators often sacrifice deterministic multiplayer synchronization for aerodynamic complexity. In Aetherfall, the requirement was uncompromised: simulate real-time lift, drag polar equations, and supersonic shockwave displacement for up to 32 players in a 64km³ volume—all while sustaining a rock-solid 120 FPS framerate with zero garbage collection allocations in the hot loop.
Standard Unity Rigidbody physics could not sustain 32 concurrent aerodynamic aircraft with multi-wing chord calculations without dropping below 45 FPS on base hardware.
Production Metrics & Constraints
02 / Custom Aerodynamic Solver Architecture
We completely bypassed the default PhysX pipeline and built a native unmanaged aerodynamic solver utilizing the Unity Job System and Burst Compiler. Each aircraft is modeled as an interconnected tree of aerodynamic lifting surfaces (wings, canards, elevators, vertical stabilizers, and fuselage camber).
By transforming aerodynamic coefficients into SIMD-friendly vector structs (using Unity.Mathematics.float4 and float8), the solver calculates Reynolds numbers, angle-of-attack induced drag, and ground effect cushions in under 0.32 milliseconds per tick across all 32 aircraft combined.
Simulation & Rollback Architecture
Entity Component & Memory Layout
Burst-Compiled Lifting Line Calculation
[BurstCompile(FloatMode = FloatMode.Fast, FloatPrecision = FloatPrecision.Standard, OptimizeFor = OptimizeFor.Performance)]
public struct LiftDragEvaluationJob : IJobParallelFor
{
[ReadOnly] public NativeArray<WingSurfaceData> Surfaces;
[ReadOnly] public NativeArray<AircraftKinematics> Kinematics;
[WriteOnly] public NativeArray<AerodynamicForceOutput> Forces;
public void Execute(int index)
{
var surface = Surfaces[index];
var state = Kinematics[surface.AircraftIndex];
// Relative airspeed in chord local space
float3 localAirVelocity = state.Velocity - state.WindVector;
float speedSq = math.lengthsq(localAirVelocity);
if (speedSq < 0.001f) return;
float dynamicPressure = 0.5f * state.AirDensity * speedSq;
float alpha = math.atan2(-localAirVelocity.y, localAirVelocity.z); // Angle of Attack
// Non-linear stall curve interpolation
float cl = math.sin(2f * alpha) * surface.LiftSlopeFactor;
float cd = surface.ZeroLiftDrag + (cl * cl) / (math.PI * surface.AspectRatio * surface.OswaldEfficiency);
float3 liftDir = math.normalize(math.cross(math.cross(localAirVelocity, surface.SpanAxis), localAirVelocity));
float3 dragDir = -math.normalize(localAirVelocity);
Forces[index] = new AerodynamicForceOutput {
ForceVector = (liftDir * cl + dragDir * cd) * dynamicPressure * surface.Area,
ApplicationPoint = surface.CenterOfPressureWorld
};
}
}03 / Volumetric Raymarched Atmospheric Shading
To deliver visceral immersion without killing the GPU budget, we implemented a custom single-pass raymarching compute shader running at half-resolution with temporal reprojection. Optical depth is sampled through a 3D Perlin-Worley noise texture, with Henyey-Greenstein phase scattering providing blinding forward silver-lining highlights when banking towards the sun.
"The flight dynamics in Aetherfall feel razor sharp. You feel the bite of the air as you pull high-G turns above the cloud deck."
— Technical Director, Polygon Arcade
04 / Hard Problems: Large-Scale Coordinate Drift
When navigating across a 64 kilometer boundary, 32-bit single-precision floating point numbers begin losing sub-millimeter precision, causing micro-jitter in camera position and cockpit instrument dials. We solved this with a hierarchical Floating Origin System: whenever the camera strays more than 2,000 meters from world (0,0,0), the entire scene graph performs a seamless world-space translation. Rigidbodies, particle origins, and physics colliders update in a single sync barrier with zero visible hitching.
05 / Key Takeaways & What I'd Do Differently
While the Burst-compiled physics solver yielded astonishing performance, developing custom spatial debugging visualizers early in the prototyping phase would have saved at least three weeks of tuning. For future projects, I would integrate continuous automated physics telemetry in CI to catch regression in aerodynamic lift curves before commits merge.