SirStone/SirRoboGarage
1
0
Fork 0
Code Issues 28 Pull Requests Actions Projects Releases Wiki Activity

feat(PPO_Bot): bot-relative bullets + scan staleness (STATE_DIM=57)

Browse Source 
- Bullet state (indices 44-55): enemy-relative → bot-relative frame
  (bot needs threat vectors to itself for dodging, not to enemy)
- New index 56: scan staleness = min(ticksSinceLastScan / 30, 1.0)
  (gives policy a confidence signal for enemy data freshness)
- warm_start.py updated: 44→57 dim expansion, TARGET_DIM variable
- Tests updated for new state layout

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
This commit is contained in:
Davide Cappellini
2026-08-20 14:27:28 +02:00
parent 6ad51148f4
commit 12624d3069
7 changed files with 718 additions and 16 deletions
Download Patch File Download Diff File Expand all files Collapse all files
PPO_Bot/state_vector.nim
+10 -8
View File
@@ -1,4 +1,4 @@
## State vector builder — produces 56-float normalized tensor for PPO policy.
## State vector builder — produces 57-float normalized tensor for PPO policy.
## No bot API imports; takes plain BotState + EnemyTracker structs.
import std/math
@@ -26,10 +26,11 @@ proc buildStateVector*(bot: BotStateData; enemy: EnemyTracker;
remainingGunAngle: float64 = 0.0;
bullets: openArray[BulletData] = [];
bulletCount: int = 0): Tensor[float32] =
## Build the 56-float normalized state tensor.
## Build the 57-float normalized state tensor.
## Indices 0-43: existing features. Indices 44-55: up to 3 bullet slots (4 floats each).
## Index 56: scan staleness (ticksSinceLastScan / 30, clamped to 1).
## All values clipped to roughly [-1, 1] via division by physical maxima.
result = zeros[float32](56)
result = zeros[float32](57)
let aW = bot.arenaWidth
let aH = bot.arenaHeight
@@ -102,14 +103,12 @@ proc buildStateVector*(bot: BotStateData; enemy: EnemyTracker;
# --- Bullet tracking (indices 44-55): up to 3 enemy bullets, 4 floats each ---
# Per bullet: relX/aW, relY/aH, speed/20, ticksToImpact/diag
# Positions are relative to enemy (threat vector). Slots beyond bulletCount stay 0.
let ex = if enemy.hasContact: enemy.current.x else: bot.arenaWidth / 2.0
let ey = if enemy.hasContact: enemy.current.y else: bot.arenaHeight / 2.0
# Positions are relative to bot (useful for dodging). Slots beyond bulletCount stay 0.
for i in 0 ..< min(bulletCount, 3):
let b = bullets[i]
let bSpeed = 20.0 - 3.0 * b.power # Tank Royale bullet speed formula
let bdx = b.x - ex
let bdy = b.y - ey
let bdx = b.x - bot.x
let bdy = b.y - bot.y
let bdist = sqrt(bdx * bdx + bdy * bdy)
let ticks = if bSpeed > 0.0: bdist / bSpeed else: 0.0
let base = 44 + i * 4
@@ -117,3 +116,6 @@ proc buildStateVector*(bot: BotStateData; enemy: EnemyTracker;
result[base + 1] = float32(bdy / bot.arenaHeight)
result[base + 2] = float32(bSpeed / 20.0)
result[base + 3] = float32(ticks / diag)
# --- Scan staleness (index 56) ---
result[56] = float32(min(enemy.current.ticksSinceLastScan.float64 / 30.0, 1.0))