Decide input encoding for enemy velocity #162

New Issue

2026-09-13T20:27:58+02:00

SirStone commented

2026-09-13 20:27:58 +02:00

Question

How should enemy velocity be encoded and fed to the SNN alongside the current bearing input?

Current input: 36 population-coded neurons for relative bearing (-180°..+180°, 10° bands).

Options for velocity:

Population-code velocity direction (another 36 neurons) + population-code speed (N neurons for speed bands)
Population-code predicted-future-bearing directly (36 neurons, same encoding as current bearing but lead-adjusted)
Cartesian velocity components (vx, vy) as population-coded channels

Design constraint: the input interface must be structured so recurrent features can augment or replace velocity features in step 3 without rewriting the SNN core.

## Question How should enemy velocity be encoded and fed to the SNN alongside the current bearing input? Current input: 36 population-coded neurons for relative bearing (-180°..+180°, 10° bands). Options for velocity: 1. Population-code velocity direction (another 36 neurons) + population-code speed (N neurons for speed bands) 2. Population-code predicted-future-bearing directly (36 neurons, same encoding as current bearing but lead-adjusted) 3. Cartesian velocity components (vx, vy) as population-coded channels Design constraint: the input interface must be structured so recurrent features can augment or replace velocity features in step 3 without rewriting the SNN core.

SirStone added the wayfinder:grilling label 2026-09-13 20:27:58 +02:00

SirStone commented

2026-09-13 20:28:07 +02:00

Blocked by: #160 (Research bullet mechanics — need to know what velocity info matters)

**Blocked by:** #160 (Research bullet mechanics — need to know what velocity info matters)

SirStone referenced this issue

2026-09-13 20:28:09 +02:00

SNNBot: lead-predicted aiming against moving targets #159

SirStone commented

2026-09-13 20:52:31 +02:00

Resolution

Polar encoding: velocity direction (36 neurons) + speed (8 neurons), 1-tick delta, SNN smooths internally.

Input vector layout (80 neurons total)

Range	Channel	Neurons	Encoding
0–35	Relative bearing	36	10° population bands, triangular interpolation (existing)
36–71	Velocity direction	36	Same 10° population scheme as bearing
72–79	Speed	8	1 unit/tick bands (0–8 range), triangular interpolation

Key decisions

Raw velocity, not pre-computed lead bearing — SNN learns the velocity→aim relationship itself; keeps the door open for recurrence in step 3
Polar, not Cartesian — matches the SNN's existing angular representation; direction uses identical encoding scheme as bearing
1-tick position delta for velocity — LIF membrane leak (τ=0.9) + 10-tick inference window handles smoothing; no preprocessing filter
Hidden layer stays at 12 — weight count goes from 432 to 960; revisit if convergence struggles
Architecture constraint met: velocity channels are additive inputs; recurrence can augment or replace them in step 3 without touching SNN core (hidden→output unchanged)

## Resolution **Polar encoding: velocity direction (36 neurons) + speed (8 neurons), 1-tick delta, SNN smooths internally.** ### Input vector layout (80 neurons total) | Range | Channel | Neurons | Encoding | |-------|---------|---------|----------| | 0–35 | Relative bearing | 36 | 10° population bands, triangular interpolation (existing) | | 36–71 | Velocity direction | 36 | Same 10° population scheme as bearing | | 72–79 | Speed | 8 | 1 unit/tick bands (0–8 range), triangular interpolation | ### Key decisions - **Raw velocity, not pre-computed lead bearing** — SNN learns the velocity→aim relationship itself; keeps the door open for recurrence in step 3 - **Polar, not Cartesian** — matches the SNN's existing angular representation; direction uses identical encoding scheme as bearing - **1-tick position delta for velocity** — LIF membrane leak (τ=0.9) + 10-tick inference window handles smoothing; no preprocessing filter - **Hidden layer stays at 12** — weight count goes from 432 to 960; revisit if convergence struggles - **Architecture constraint met**: velocity channels are additive inputs; recurrence can augment or replace them in step 3 without touching SNN core (hidden→output unchanged)

SirStone closed this issue

2026-09-13 20:52:32 +02:00

SirStone referenced this issue

2026-09-13 20:52:38 +02:00

SNNBot: lead-predicted aiming against moving targets #159

Sign in to join this conversation.

1 Participants

Notifications

Due Date

No due date set.

Dependencies

No dependencies set.

Reference: SirStone/SirRoboGarage#162