A NEAT-based ML agent trained through curriculum learning.
Demonstrating the power of neuroevolution in the Resonance Engine.
NEAT neuroevolution agent trained through curriculum learning
Pure Rust implementation of NeuroEvolution of Augmenting Topologies. Evolves neural network topology and weights simultaneously.
10-dimensional normalized state vectors: player position, velocity, goal location, hazard detection, and wall proximity.
6 discrete actions: noop, move left, move right, jump, move left + jump, move right + jump.
Progressive difficulty: L1 (Move Right) → L2 (Jump to Goal) → L3 (Avoid Enemy) → L4 (Platform Climb) → L5 (Gauntlet).
100% success on training levels; generalization to novel variants in progress
| Level | Training | Generalization | Status |
|---|---|---|---|
| L1: Move Right | 100% | 100% | ✓ Complete |
| L2: Jump to Goal | 100% | 35% | In Progress |
| L3: Avoid Enemy | 100% | 45% | In Progress |
| L4: Platform Climb | 100% | TBD | In Progress |
| L5: Gauntlet | 100% | TBD | In Progress |
NEAT neuroevolution with curriculum learning
# Clone the repository (when available) # cd resonance-agent # Train Rezzy on platformer challenges cargo run --release --bin train_platformer # Test a trained model cargo run --release --bin test_model -- --model models/level5_gauntlet.json # View training results cargo run --release --bin view_stats
Explore the engine and NAL that power Rezzy's learning capabilities.