This project implements an intelligent traffic flow optimization system using Deep Reinforcement Learning (DRL) techniques. The system adaptively controls vehicle speed and lane changes to enhance traffic efficiency, reduce congestion, and improve road safety.
The project develops and compares two Reinforcement Learning approaches:
- Deep Q-Network (DQN)
- Actor-Critic Network
Both agents learn to optimize traffic flow by making real-time decisions about vehicle speed and lane changes based on the surrounding traffic conditions.
- Real-time vehicle control optimization
- Multi-agent reinforcement learning
- Adaptive speed regulation
- Intelligent lane change decisions
- Safety-aware decision making
- High-frequency state updates (10 Hz)
The environment state consists of 8 key features:
| Feature | Description | Unit |
|---|---|---|
| Vehicle Speed | Current velocity | m/s |
| Vehicle Acceleration | Rate of speed change | m/s² |
| Lane Position | Current lane identifier | - |
| Space Headway | Distance to preceding vehicle | m |
| Time Headway | Time gap to preceding vehicle | s |
| Vehicle Class | Type of vehicle | - |
| Global X | X-coordinate in global frame | m |
| Global Y | Y-coordinate in global frame | m |
The agent can perform 5 distinct actions:
| Action | Description | Conditions | Change |
|---|---|---|---|
| 0 | Maintain speed | No conditions | 0 m/s |
| 1 | Increase speed | Space_Headway ≥ 15m | +2 m/s |
| 2 | Decrease speed | Space_Headway < 10m | -2 m/s |
| 3 | Change left | Left lane exists & unoccupied | Lane shift |
| 4 | Change right | Right lane exists & unoccupied | Lane shift |
- Safe Following Distance: Minimum 15 meters from preceding vehicle
- Collision Risk Zone: Less than 5 meters gap
- Target Speed: 27 m/s (approximately 60 mph)
The reward function balances optimal speed maintenance with collision avoidance:
R = (10 - |V_t - V_optimal|) - P_collision
Where:
- V_t = Current vehicle speed (m/s)
- V_optimal = 27 m/s (target speed)
- P_collision = Penalty factor
- 20 if Space_Headway < 5m (high collision risk)
- 0 otherwise
- TensorFlow/Keras for DRL implementation
- NumPy for numerical computations
- Pandas for data handling
- Matplotlib for visualization
- High-frequency trajectory data (10 Hz)
- Each frame represents 0.1-second interval
- Rich vehicle state information
- Clone the repository:
git clone https://github.com/yourusername/Reinforcement-Learning-based-Traffic-Optimization.git- Install required dependencies:
pip install tensorflow numpy pandas matplotlib- Run the Jupyter notebook:
jupyter notebook "Traffic_Flow_Optimization final.ipynb"The project demonstrates:
- Improved traffic flow efficiency
- Reduced congestion through intelligent speed control
- Enhanced safety through collision avoidance
- Adaptive behavior in various traffic conditions
Detailed documentation is available in:
Traffic_Flow_Optimization final.ipynb- Main implementation notebookGroup_111-TrafficOptimization.pdf- Project documentation
Contributions are welcome! Please feel free to submit a Pull Request.
This project is licensed under the MIT License - see the LICENSE file for details.
- Dataset provided by [Dataset Source]
- Based on research in Deep Reinforcement Learning for traffic optimization
- Inspired by modern traffic management systems