Initial commit — Elden Ring RL agent

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

elden_env.py

@@ -0,0 +1,95 @@
+import gymnasium as gym
+from gymnasium import spaces
+import numpy as np
+from vision_agent import EldenVision
+from input_driver import EldenController
+import time
+from evdev import ecodes as e
+
+class EldenRingEnv(gym.Env):
+    def __init__(self):
+        super(EldenRingEnv, self).__init__()
+        self.vision = EldenVision(stack_size=4)
+        self.controller = EldenController()
+        self.observation_space = spaces.Box(low=0, high=255, shape=(640, 640, 12), dtype=np.uint8)
+        self.action_space = spaces.MultiDiscrete([4, 5, 3, 3, 2])
+        
+        self.last_p_hp = 100.0
+        self.last_b_hp = 100.0
+        self.frame_skip = 4 
+
+    def step(self, actions):
+        combat, move, cam_x, cam_y, switch = actions
+        if switch == 1: self.controller.switch_item()
+        if combat == 1: self.controller.dodge()
+        elif combat == 2: self.controller.attack()
+        elif combat == 3: self.controller.use_item()
+        self.controller.move(move)
+        
+        mx = -50 if cam_x == 1 else (50 if cam_x == 2 else 0)
+        my = -50 if cam_y == 1 else (50 if cam_y == 2 else 0)
+        if mx != 0 or my != 0: self.controller.look(mx, my)
+
+        for _ in range(self.frame_skip):
+            obs, hp, fp, sp, boss = self.vision.get_state()
+
+        reward = 0.01
+        if combat == 3 and hp > 90: reward -= 5.0
+        if hp < self.last_p_hp: reward -= 20.0
+        if boss < self.last_b_hp: reward += 10.0 * ((self.last_b_hp - boss) / 5.0)
+
+        self.last_p_hp, self.last_b_hp = hp, boss
+        done = hp <= 0 or (boss <= 0 and self.last_b_hp > 10)
+        return obs, reward, done, False, {}
+
+    def reset(self, seed=None, options=None):
+        super().reset(seed=seed)
+        print(f"\n[NAV] DEATH DETECTED. Waiting 25s for Respawn...")
+        time.sleep(25) 
+        
+        print("[NAV] Running to Maliketh Fog Gate...")
+        
+        # 1. Exit Grace Room
+        self.controller.move(1) # W
+        time.sleep(2.0)
+        
+        # 2. Turn Right to stairs
+        self.controller.look(1800, 0) 
+        time.sleep(0.5)
+        
+        # 3. Run up the stairs
+        self.controller.move(1) 
+        time.sleep(4.5)
+        
+        # 4. Turn Left to bridge
+        self.controller.look(-1600, 0)
+        time.sleep(0.5)
+        
+        # 5. Sprint across bridge and look for fog
+        print("[NAV] Sprinting across bridge...")
+        start_nav = time.time()
+        found_fog = False
+        
+        while time.time() - start_nav < 25: # 25s timeout for navigation
+            with self.vision.lock:
+                hsv = self.vision.frame_hsv.copy()
+            
+            if self.vision.detect_fog(hsv):
+                print("[NAV] FOG DETECTED! Entering Arena...")
+                self.controller.move(0) 
+                time.sleep(0.2)
+                self.controller.press(e.KEY_E) # Interact
+                time.sleep(5.0) # Extra safety for entry animation
+                found_fog = True
+                break
+                
+            self.controller.move(1) # Keep running forward
+            time.sleep(0.1)
+
+        if not found_fog:
+            print("[NAV] Warning: Fog not found. Sentinel might have blocked us.")
+        
+        self.controller.move(0)
+        obs, hp, fp, sp, boss = self.vision.get_state()
+        self.last_p_hp, self.last_b_hp = hp, boss
+        return obs, {}