csd4ni3l/fleet-commander
1
0
Fork 0
mirror of https://github.com/csd4ni3l/fleet-commander.git synced 2026-01-01 04:23:47 +01:00
Code Issues Packages Projects Releases 1 Wiki Activity

add better model, improve rewards

Browse Source
This commit is contained in:
csd4ni3l
2025-11-17 15:44:42 +01:00
parent c7c22695e5
commit 6e18ac3ae4
3 changed files with 5 additions and 4 deletions
Download Patch File Download Diff File Expand all files Collapse all files

3
.gitignore vendored
View File

@@ -180,4 +180,5 @@ test*.py
logs/ logs/
logs logs
settings.json settings.json
training_logs training_logs
*.bck*

BIN
invader_agent.zip Normal file
View File

Binary file not shown.

6
utils/rl.py
View File

@@ -178,7 +178,7 @@ class SpaceInvadersEnv(gym.Env):
nearest = self._nearest_enemy() nearest = self._nearest_enemy()
alignment = abs(nearest.center_x - self.player.center_x) / self.width alignment = abs(nearest.center_x - self.player.center_x) / self.width
if alignment < 0.025: if alignment < 0.025:
reward += 0.1 reward += 0.005
self.player.center_x = np.clip(self.player.center_x, 0, self.width) self.player.center_x = np.clip(self.player.center_x, 0, self.width)
self.player_speed = (self.player.center_x - prev_x) / max(1e-6, PLAYER_SPEED) self.player_speed = (self.player.center_x - prev_x) / max(1e-6, PLAYER_SPEED)
@@ -260,11 +260,11 @@ class SpaceInvadersEnv(gym.Env):
self.bullets.append(b) self.bullets.append(b)
if self.player_alive: if self.player_alive:
edge_threshold = self.width * 0.15 edge_threshold = self.width * 0.1
if self.player.center_x < edge_threshold or self.player.center_x > self.width - edge_threshold: if self.player.center_x < edge_threshold or self.player.center_x > self.width - edge_threshold:
reward -= 0.03 reward -= 0.03
reward -= 0.0025 reward -= 0.01
obs = self._obs() obs = self._obs()