Notes on 1D Nonlinear Transformations for Games

game-dev
notes
My notes from Squirrel Eiserloh’s presentation on 1D nonlinear transformations for game development.
Author

Christian Mills

Published

December 29, 2021

Overview

Here are some notes I took while watching Squirrel Eiserloh’s presentation covering how 1D nonlinear transformations can be used by game programmers.

Motivations

Implicit versus Parametric Equations

  • Implicit equations are rules:
    • Equation for a circle: \(x^{2} + Y^{2} = 25\)
      • A point is either on the circle or not
  • Parametric functions
    • Yield an output for an input value
      • \(P_{x} = 5 \cdot cos(2 \pi \cdot t)\)
      • \(P_{y} = 5 \cdot sin(2 \pi \cdot t)\)
    • \(P(t) = ?\)
    • \(P(t) = (t, t \cdot cos(t), t*sin(t))\)
      • \((x, y, z)\)
      • Generates a spiral that increases in radius along the x axis
    • Anything you can express in terms of a single float as input
    • A common float input is “time”

Parametric Manipulations

  • Do NOT mess with the interpolation itself (e.g. color, position, AI disposition, etc.)
  • Instead just mess the parameter

Parametric Opportunities

  • Anytime you have a single float to change
  • Anytime you can express something in terms of a single float
  • Pretty much whenever you use time

The Big Idea

  • You can make any parametric equation more interesting without modifying the function itself, without knowing anything about the function

The Two Most Important Number Ranges

  • \([0,1]\)
    • Useful for fractions
      • % shadow
      • % luminance
      • % falloff
      • % complete
      • % damage
      • % experience
      • % cost
      • % penalty
      • % fog
      • % AI aggression
      • % chance to hit
      • % chance to drop loot
      • % time to complete
      • Fuzzy Logic
      • Most anything parametric
  • \([-1,1]\)
    • Useful for deviations
      • noise
      • perturbation
      • terrain and map generation
      • variation
      • distribution
      • sinusoidal
      • AI response curves

Normalized Non-Linear Functions

  • \([0,1]\)
  • Functions for which:
    • \(P(0) = 0\)
    • \(P(1) = 1\)
    • \(P(t) \ != t\)
  • Examples
    • Position over time
    • Scale over time
    • Alpha over time
    • Color over time
    • Strength over time
    • Aggression over time
  • Also called
    • easing functions
    • filter functions
    • lerping functions
    • tweening functions

Range Mapping

  • can be applied during middle of range-mapping
out RangeMap(in, inStart, inEnd, outStart, outEnd)
{
    // Puts in [0, inEnd - inStart]
    out = in - inStart;
    // Puts in [0,1]
    out /= (inEnd - inStart);
    // in [0,1]
    out = ApplySomeEasingFunction(out);
    // Puts in [0, outRange]
    out *= (outEnd - outStart);
    // Puts in [outStart, outEnd]
    return out + outStart
}

SmoothStart

  • \(SmoothStartN(t) = t^{n}\)
  • Larger exponents result in steeper curve
  • Will always start and end at the same time, regardless of exponent value
  • Technique
    • exponentiating

SmoothStop

  • \(SmoothStopN(t) = 1 - (1 - t)^{n}\)
  • Larger exponents results in longer braking period at the end
  • Techniques
    • exponentiating
    • flipping

\(Mix(a, b, weightB, t)= a + weightB(b-a)\)

  • \(Mix(SmoothStart2, SmoothStop2, blend, t)\)
  • \(SmoothStart2.2 = Mix(SmoothStart2, SmoothStart3, 0.2);\)
    • Way faster than using the pow() function

Crossfade

  • Like Mix, but use t itself as the mix weight
  • Also called SmoothStep

Scale

  • \(Scale(Function, t) = t \cdot Function(t)\)

ReverseScale

  • \(ReverseScale(Function, t) = (1-t) \cdot Function(t)\)

\(Arch2(t) = Scale(Flip(t)) = t \cdot (1-t)\)

\(SmoothStartArch3(t) = Scale(Arch2, t) = t^{2}(1-t)\)

\(SmoothStopArch3(t) = ReverseScale(Arch2, t) = t(1-t)^{2}\)

\(SmoothStepArch3(t) = ReverseScale(Scale(Arch2, t), t)\)

\(BellCurve6(t) = SmoothStop3(t) \cdot SmoothStart3(t)\)