Modern Autonomous Driving System

An autonomous driving system using YOLOv8 for object detection and semantic segmentation for lane detection in the CARLA simulator.

Features

• YOLOv8 object detection for vehicles, pedestrians, and traffic signs
• Semantic segmentation for precise lane detection
• Intelligent safety assessment and emergency response
• Adaptive vehicle control with object awareness
• GPU acceleration support
• Real-time performance monitoring

Prerequisites

• Python 3.8+
• CARLA Simulator 0.9.13+
• NVIDIA GPU with CUDA support (recommended)

Installation

Quick Setup

python setup_modern_cv.py

Manual Setup

pip install -r requirements.txt
mkdir models data logs outputs

Usage

1. Start CARLA simulator
2. Run the system:

   python src/main.py

Controls

• 'q' or ESC: Quit
• 'r': Reset controllers
• 's': Show statistics
• 'p': Performance metrics
• 'h': Help

System Architecture

The system consists of three main components:

1. ModernLaneDetector: YOLOv8 object detection and UNet semantic segmentation
2. ModernVehicleController: Adaptive PID control with safety features
3. CarlaConnector: Enhanced CARLA interface with robust error handling

Configuration

Model Selection

• yolov8n: Fastest, lowest accuracy
• yolov8s: Balanced performance
• yolov8m: Better accuracy
• yolov8l: High accuracy
• yolov8x: Best accuracy

Performance Tuning

# Adjust in lane_detection.py
lane_detector.yolo_model = YOLO('yolov8s.pt')

# Adjust in vehicle_controller.py
vehicle_controller.kp_steer = 0.003
vehicle_controller.base_speed = 35.0

Technical Details

Object Detection

• Vehicles: cars, motorcycles, buses, trucks
• Pedestrians: people detection
• Traffic signs: stop signs, traffic lights

Lane Segmentation

• Ego lane: current driving lane
• Adjacent lanes: neighboring lanes
• Road boundaries: road edges
• Drivable area: safe driving regions

Safety Features

• Emergency braking for critical situations
• Adaptive speed control based on safety assessment
• Object avoidance steering adjustments
• Continuous lane center tracking

Troubleshooting

CARLA Connection Issues:

• Ensure CARLA simulator is running
• Check host/port configuration

Performance Issues:

• Use smaller YOLO model (yolov8n)
• Enable GPU acceleration
• Reduce image resolution

Detection Issues:

• Verify model files are downloaded
• Check lighting conditions in CARLA
• Adjust confidence thresholds

Development

The system is modular and extensible. Key files:

• src/lane_detection.py: Computer vision pipeline
• src/vehicle_controller.py: Vehicle control logic
• src/carla_connector.py: CARLA interface
• src/main.py: Main application loop

Status

This is a work in progress implementing modern computer vision techniques for autonomous driving research and development.