Simplest ai trick gdc2013 dino v2 (1)

Simplest AI Trick in the BookNormalised Tunable Sigmoid Function

Dino Dini NHTV University of Applied Sciences

Normalized Values Are Useful

For example:

● Utility calculations

● Input management

● Control systems

● Tunable parameters

Analog Input

Abstract away the device dependent positional values

(0 to 255? -1024 to 1024?) and normalise.

Normalised values are much easier to work with.

-1 10

Example:Analog Input - Left / Right Rotation

Input Device

1024

-1024

Movement Driver

1

-1

RotationDegrees per

Frame

Normalizer

5

-5


Input Device

1024

-1024

Movement Driver

1

-1

RotationDegrees per

Frame

Normalizer

5

-5

Linear relationship

Degrees rotation per frame

Control input (left - right)

Linear relationship



I want greater sensitivity

Linear relationship



Linear relationship



I also want full range

Linear relationship



Greater sensitivity

Full Range


Input Device

1024

-1024

Movement Driver

1

-1

RotationDegrees per

Frame

Normalizer

5

-5

Sigmoid like

function

1

-1


Input Device

1024

-1024

Movement Driver

1

-1

RotationDegrees per

Frame

Normalizer

5

-5

Sigmoid like

function

1

-1

k

Sigmoid function?

Logit function?

Normalised Tunable (half) Sigmoid Function?


k = 0.2


k = 0.01


k = 2


k = -1.2


k = -1.01


k = -3

Normalised Tunable Sigmoid Function

k = 0.2

Normalised Tunable Sigmoid Function

Thank you

Simplest ai trick gdc2013 dino v2 (1)

Documents

normalised values

logit function

ksigmoid function

right rotationinput

movement driver1

usefulfor example

simplest ai trick