Environment Guide¶

Complete guide to using the EldenGym environment.

Creating an Environment¶

import gymnasium as gym
import eldengym

env = gym.make("EldenGym-v0")

Environment Lifecycle¶

1. Reset¶

Start a new episode:

observation, info = env.reset()

print(f"Player HP: {info['player_hp']}")
print(f"Boss HP: {info['target_hp']}")
print(f"Observation shape: {observation.shape}")

2. Step¶

Execute an action:

action = env.action_space.sample()
obs, reward, terminated, truncated, info = env.step(action)

if terminated:
    print("Episode ended (boss defeated or player died)")
if truncated:
    print("Episode truncated (max steps reached)")

3. Render¶

Get the current frame:

frame = env.render()
# Returns RGB numpy array (H, W, 3)

4. Close¶

Clean up resources:

env.close()

Complete Example¶

import gymnasium as gym
import eldengym

# Create environment
env = gym.make("EldenGym-v0", scenario_name="margit")

# Training loop
for episode in range(10):
    obs, info = env.reset()
    episode_reward = 0

    done = False
    while not done:
        # Your agent logic here
        action = env.action_space.sample()

        obs, reward, terminated, truncated, info = env.step(action)
        episode_reward += reward
        done = terminated or truncated

        # Optionally render
        # frame = env.render()

    print(f"Episode {episode}: Reward = {episode_reward:.2f}")

env.close()

Configuration Options¶

See the Configuration Guide for all options.

Common Patterns¶

Fixed Episode Length¶

env = gym.make("EldenGym-v0", max_step=1000)

obs, info = env.reset()
for step in range(1000):
    action = env.action_space.sample()
    obs, reward, terminated, truncated, info = env.step(action)
    if terminated or truncated:
        break

Monitoring Stats¶

obs, info = env.reset()
stats = {
    'player_hp': [],
    'target_hp': [],
    'rewards': []
}

for step in range(1000):
    action = agent.predict(obs)
    obs, reward, terminated, truncated, info = env.step(action)

    stats['player_hp'].append(info['player_hp'])
    stats['target_hp'].append(info['target_hp'])
    stats['rewards'].append(reward)

    if terminated or truncated:
        break

# Plot stats
import matplotlib.pyplot as plt
plt.plot(stats['player_hp'], label='Player HP')
plt.plot(stats['target_hp'], label='Boss HP')
plt.legend()
plt.show()

Custom Reset Logic¶

class CustomEnv(gym.Wrapper):
    def reset(self, **kwargs):
        obs, info = super().reset(**kwargs)
        # Custom initialization
        self.env.unwrapped.client.set_player_hp(9999)
        return obs, info

env = CustomEnv(gym.make("EldenGym-v0"))

Troubleshooting¶

Environment Won't Reset¶

Problem: Reset hangs or fails

Solutions: - Ensure Siphon server is running - Check game is at the correct location - Verify memory patterns are correct

Actions Don't Work¶

Problem: Actions have no effect in game

Solutions: - Ensure input subsystem is initialized - Check game has focus - Verify key mappings match your game settings

Poor Performance¶

Problem: Environment runs slowly

Solutions:

# Increase frame skip
env = gym.make("EldenGym-v0", frame_skip=8)

# Disable frame capture if not needed
env = gym.make("EldenGym-v0", observation_mode="dict")

# Adjust game speed
env = gym.make("EldenGym-v0", game_speed=2.0)