Autonomous Vehicle Simulation Environment

Recreation of the simulation environment from the paper: "Safety and Risk Analysis of Autonomous Vehicles Using Computer Vision and Neural Networks" by Dixit et al. (2021)

Overview

This project recreates the three main case studies from the paper:

1. Straight Lane Detection using Hough Line Transform
2. Curved Lane Detection using OpenCV and HSV color space
3. Behavioral Planning with Model-based Reinforcement Learning and Robust Control

Project Structure

├── av_simulation.py          # Main simulation with 3 environments (highway, merge, roundabout)
├── lane_detection.py          # Lane detection algorithms (straight & curved)
├── behavioral_planning.py     # MDP, MRL, and robust control implementation
├── requirements.txt          # Python dependencies
└── README.md                # This file

Features

1. Simulation Environments (av_simulation.py)

Based on Case Study 3, implements three driving scenarios:

• Highway Environment: 4-lane highway with multiple vehicles
• Lane Merging: Highway with service road merge point
• Roundabout: 4-way roundabout navigation

Key Parameters from Paper:

• Acceleration Range: (-5, 5.0) m/s²
• Steering Range: ±45 degrees
• Max Speed: 40 m/s
• Default Speeds: [23, 25] m/s
• Perception Distance: 180 m

2. Lane Detection (lane_detection.py)

Straight Lane Detection (Case Study 1):

• Canny edge detection with 5×5 Gaussian filter
• Region of Interest (ROI) segmentation
• Hough Line Transform for lane marking detection
• Line averaging and extrapolation

Curved Lane Detection (Case Study 2):

• Camera distortion correction
• Perspective transformation (bird's eye view)
• HSV color space filtering for yellow/white lanes
• Sobel operator for edge detection
• 2nd-degree polynomial fitting: x = Ay² + By + C
• Radius of curvature calculation

3. Behavioral Planning (behavioral_planning.py)

Model-Based Reinforcement Learning:

• Neural network dynamics model: ẋ = f_θ(x, u) = A_θ(x, u)x + B_θ(x, u)u
• Experience buffer with 2000 training epochs
• Trajectory prediction with learned dynamics

Planning Algorithms:

• Cross-Entropy Method (CEM) for trajectory optimization
• Robust Control Framework with model uncertainty
• Continuous Ambiguity Prediction for neighboring vehicles
• Partially Observable MDP (POMDP) implementation

Installation

1. Clone or download this repository
2. Install dependencies:

pip install -r requirements.txt

Usage

Run Main Simulation

python av_simulation.py

Controls:

• 1, 2, 3 - Switch between Highway, Merge, and Roundabout environments
• SPACE - Pause/Resume
• R - Reset current environment
• Arrow Keys - Manual vehicle control (for testing)
• ESC - Exit

The simulation includes:

• Green ego vehicle with autonomous behavior planning
• Blue traffic vehicles
• Collision detection
• Real-time speed and position display

Run Lane Detection Demo

python lane_detection.py

Controls:

• s - Switch to Straight Lane Detection
• c - Switch to Curved Lane Detection
• q - Quit

The demo creates simulated road scenes and applies the detection algorithms.

Run Behavioral Planning Demo

python behavioral_planning.py

This runs a demonstration of:

1. Model training with synthetic data
2. Trajectory prediction
3. Cross-entropy optimization
4. Robust control with uncertainty
5. Continuous ambiguity prediction

Implementation Details

Vehicle Dynamics

The vehicle model uses bicycle dynamics with state vector:

• Position (x, y)
• Velocity (vx, vy)
• Heading angle
• Current lane

Behavioral Planner

The planner considers:

• Distance to front vehicle
• Safe lane change opportunities
• Speed maintenance within limits
• Collision avoidance

Robust Control

Addresses model errors from Case Study 3:

• Considers multiple possible future states
• Worst-case scenario planning
• Driving style estimation (aggressive/normal/conservative)

Key Findings Recreated

1. Lane Detection: Successfully detects both straight and curved lanes using computer vision
2. Behavioral Planning: MDP-based planning reduces collision risk
3. Robust Control: Considering uncertainty prevents accidents shown in Figure 11 of the paper

Simulation Parameters

All parameters match Table 2 from the paper:

Parameter	Value
Acceleration Range	(-5, 5.0) m/s²
Steering Range	(-0.785, 0.785) rad
Max Speed	40 m/s
Default Speeds	[23, 25] m/s
Distance Wanted	10.0 m
Time Wanted	1.5 s
Perception Distance	180 m

Limitations

This is a simplified recreation for educational purposes:

• Uses pygame instead of actual vehicle hardware
• Synthetic data instead of real LIDAR/camera feeds
• Simplified physics model
• No actual V2X communication

Future Improvements

• Add pedestrian and cyclist models
• Implement V2X communication simulation
• Add more complex road scenarios
• Include weather conditions (fog, rain)
• Implement full SCNN for better spatial feature detection

References

Dixit, A., Kumar Chidambaram, R., & Allam, Z. (2021). Safety and Risk Analysis of Autonomous Vehicles Using Computer Vision and Neural Networks. Vehicles, 3, 595-617.

License

This is an educational recreation of academic research. Please refer to the original paper for scientific details.