The AI Simulation Project is a cutting-edge environment designed to simulate and evolve AI models using evolutionary neural networks. With highly customizable parameters and advanced visualization, this project allows users to analyze AI decision-making, observe adaptive behaviors, and optimize neural architectures efficiently.
- 🧠 Neural Network Decision Making – Supports both TensorFlow and NumPy implementations.
- 🔄 Evolutionary Algorithms – Adaptive learning through natural selection and genetic evolution.
- ⚡ Performance Optimization – NumPy fast path optimizations for enhanced speed.
- 🎨 Advanced Visualization – Real-time tracking of AI learning and fitness progression.
- 🛠️ Customizable Parameters – Adjust settings via command-line arguments.
- 🏆 Benchmarking & Statistics – Collect performance metrics and optimize models.
- 💾 Checkpoint System – Save and load progress automatically or manually.
- ⚡ Headless Mode Support – Run high-speed training without GUI overhead.
The AI Dot Collector is a core simulation showcasing evolutionary AI in action. In this scenario, dots learn to navigate toward a target while avoiding obstacles, improving their strategy across multiple generations.
✅ Neural Network-Based Movement – Each dot makes decisions using a trained neural model.
✅ Full 360° Navigation – Control speed, direction, and acceleration dynamically.
✅ Enhanced Fitness Calculation – Efficient navigation is rewarded; stagnation and collisions are penalized.
✅ Evolutionary Learning – Populations evolve over multiple generations.
✅ Real-time Stats & Graphs – Track AI learning curves, fitness history, and progress.
✅ Save & Load AI Brains – Preserve trained models for future analysis.
✅ Optimized Sensor Calculations – Fast, precise environmental detection for each AI agent.
✅ Comprehensive Error Handling – Ensures smooth execution with minimal failures.
Clone the repository and install dependencies:
git clone https://github.com/Mahdiglm/AI-Simulation.git
cd AI-Simulation
pip install -r requirements.txtRun the simulation with customizable options:
python simulation.py [options]--no-tensorflow # Use pure NumPy neural networks
--no-fast-path # Disable NumPy fast path optimization
--no-gui # Run in headless mode (no GUI)
--benchmark # Run performance benchmark
--train N # Train for N generations
--load FILE # Load checkpoint from a saved file Run the simulation with default settings:
python simulation.pyTrain AI for 100 generations in headless mode:
python simulation.py --train 100 --no-guiLoad a saved checkpoint:
python simulation.py --load checkpoints/checkpoint_gen_50.jsonRun a performance benchmark:
python simulation.py --benchmarkWhile running, control the simulation using:
| Key | Action |
|---|---|
| Space | Pause/Resume simulation |
| R | Reset simulation |
| S | Save checkpoint |
| 1 | Normal speed |
| 2 | 2x speed |
| 5 | 5x speed |
| 0 | 10x speed |
| D | Toggle between best dot & all dots |
| G | Show/hide fitness graph |
📂 AI-Simulation/
├── simulation.py # Main simulation manager
├── dot.py # AI-controlled dots with sensors & movement
├── population.py # Evolutionary population management
├── neural_network.py # NumPy-based neural network
├── tf_neural_network.py # TensorFlow-based neural network
├── obstacle.py # Handles obstacles & collision detection
├── constants.py # Global constants
├── runs/ # Stored simulation data & checkpoints
-
Input Layer (15 neurons)
- 8 distance sensors (N, NE, E, SE, S, SW, W, NW)
- 2 normalized target coordinates (dx, dy)
- 2 velocity components (vel_x, vel_y)
- 2 distance metrics (nearest obstacle & target distance)
- 1 normalized speed value
-
Hidden Layers
- Configurable (default: 2 layers, 16 neurons each)
- Activation: Leaky ReLU
-
Output Layer (3 neurons)
- Direction (0 - 2π)
- Acceleration (-1 to 1)
- Speed Control (0 to 1)
- Selection – Tournament selection strategy
- Crossover – Single-point crossover for genetic recombination
- Mutation – Controlled random weight mutations
- Fitness Calculation – Distance, efficiency, progress evaluation
- Elitism – Best performers are preserved across generations
- NumPy & TensorFlow implementations for flexibility
- Optimized feedforward calculations for fast neural processing
- Efficient collision detection with early-exit optimizations
- Configurable execution speed for training vs. visualization
Simulation runs are stored in:
📂 runs/
├── run_[timestamp]/ # Each run has its own directory
│ ├── checkpoints/ # Saved AI models
│ ├── stats/ # JSON-formatted performance statistics
This project is licensed under the MIT License.
Contributions are welcome! If you'd like to improve the project, please submit a Pull Request.
💡 Enhance your AI simulation skills & explore the power of neural evolution! 🚀