Revenant NeuroSpawn is a deep RL-powered zombie spawner for shooter game engineered to make intelligent spawn decisions.
- State Simulation: Custom environment simulates player health, position, zombie activity, spawn point distances, and time since last spawn for realistic agent training.
- Running Normalizer: Continuously updates and normalizes game state inputs, ensuring stable RL training and generalization across varied scenarios.
- Strategic Reward Function: Encourages intelligent spawning—avoiding overwhelming the player, using distant spawns when health is low, and promoting diverse spawn locations when conditions are favorable.
- Actionable RL Training: Trained using policy gradients with entropy regularization and learning rate scheduling for stable convergence and diverse behaviors.
- Performance Visualization:
- Input: Normalized state vector of size 11 (player health, position (x, y, z), zombies active, 5 spawn point distances, time since last spawn).
- Shared Layers:
- Linear (11 → 256), LayerNorm, ReLU
- Linear (256 → 256), LayerNorm, ReLU
- Linear (256 → 128), ReLU
- Spawn Point Heads: For each of the 5 spawn points:
- Linear (128 → 64), ReLU
- Linear (64 → 4) — logits for spawning 0, 1, 2, or 3 zombies
- Output: Tensor (batch_size, 5, 4) — action logits for each spawn point.
This architecture enables the agent to learn both global context and specialized spawning strategies, resulting in dynamic, context-aware zombie spawning and an engaging player experience.