Input normalization strategy #75

New Issue

2026-08-25T15:07:15+02:00

SirStone commented

2026-08-25 15:07:15 +02:00

Question

What normalization strategy for the ANN's 91 inputs (30 ticks × lateral_vel, heading_delta, wall_distance)?

Parent map: #74

## Question What normalization strategy for the ANN's 91 inputs (30 ticks × lateral_vel, heading_delta, wall_distance)? Parent map: https://git.fossellini.top/SirStone/SirRoboGarage/issues/74

SirStone added the wayfinder:grilling label 2026-08-25 15:07:15 +02:00

SirStone commented

2026-08-25 15:07:32 +02:00

Resolution

Per-feature fixed scaling using game constants. Each feature divided by its known max from game physics:

lateral_vel / 8.0 → roughly [-1, 1]
heading_delta / π → [-1, 1]
wall_distance / 1000.0 → [0, 1]

Mixed ranges per feature ([0,1] and [-1,1]) — tanh doesn't need symmetric inputs, just bounded magnitude. Zero runtime cost, no running stats, deterministic. The GA compensates for any imprecision in scaling.

## Resolution **Per-feature fixed scaling using game constants.** Each feature divided by its known max from game physics: - `lateral_vel / 8.0` → roughly [-1, 1] - `heading_delta / π` → [-1, 1] - `wall_distance / 1000.0` → [0, 1] Mixed ranges per feature ([0,1] and [-1,1]) — tanh doesn't need symmetric inputs, just bounded magnitude. Zero runtime cost, no running stats, deterministic. The GA compensates for any imprecision in scaling.

SirStone closed this issue

2026-08-25 15:08:01 +02:00

SirStone referenced this issue

2026-08-25 15:08:20 +02:00

Evo_Bot implementation design decisions #74

SirStone referenced this issue

2026-08-25 15:08:24 +02:00

Evo_Bot implementation design decisions #74

Sign in to join this conversation.