CARLA-ROS2: Modular Autonomous Driving Stack

A complete autonomous driving system for the CARLA simulator, built with ROS2. Implements a full Perception → Planning → Control pipeline using modular ROS2 nodes.

What It Does

The vehicle autonomously:

• Follows lanes using camera and LiDAR perception
• Detects vehicles, pedestrians, and cyclists with YOLOv8
• Detects and obeys traffic lights (Red → Stop, Green → Go)
• Maintains safe following distance (2-second rule)
• Stops at red lights with smooth deceleration
• Avoids obstacles in the lane
• Changes lanes to overtake slow vehicles
• Reaches a destination through urban traffic (Town03)

Project Documentation & Specifications

• Implementation Plan & Architecture: Full design specifications, sensor fusion algorithms, and dataset split strategy.
• Development Tasks & Roadmap: Task checklist, completed milestones, and future features roadmap.

Architecture

                    CARLA Simulator (Docker)
                           │
              ┌────────────┴────────────┐
              │                         │
           Camera                     LiDAR
              │                         │
              ▼                         ▼
     ┌─────────────────┐     ┌──────────────────┐
     │ Perception Node │     │ LiDAR Processing │
     │   (YOLOv8)      │     │  (RANSAC+DBSCAN) │
     └────────┬────────┘     └────────┬─────────┘
              │                       │
              └───────────┬───────────┘
                          ▼
                ┌──────────────────┐
                │  Sensor Fusion   │
                │     (EKF)        │
                └────────┬─────────┘
                         ▼
                ┌──────────────────┐
                │ Object Tracking  │
                │  (ByteTrack)     │
                └────────┬─────────┘
                         ▼
                ┌──────────────────┐
                │ Decision Making  │
                │    (FSM)         │
                └────────┬─────────┘
                         ▼
                ┌──────────────────┐
                │  Path Planning   │
                │ (Hybrid A*)      │
                └────────┬─────────┘
                         ▼
                ┌──────────────────┐
                │ Vehicle Control  │
                │(PID+Pure Pursuit)│
                └────────┬─────────┘
                         ▼
                    CARLA Vehicle

Technology Stack

Technology	Purpose
Ubuntu 24.04	Operating System
ROS2 Jazzy	Robot middleware
CARLA 0.9.15	Driving simulator (Docker)
Python 3.12	Primary language
PyTorch + CUDA	Deep learning inference
YOLOv8 (Ultralytics)	Object detection
OpenCV	Image processing
NumPy / SciPy	Numerical computation
scikit-learn	DBSCAN clustering

ROS2 Packages

Package	Description	Key Algorithms
carla_ad_msgs	Custom message definitions	9 message types
carla_bridge	CARLA ↔ ROS2 interface	Sync mode, sensor spawning
perception	Camera object detection	YOLOv8n, HSV traffic light classification
lidar_processing	LiDAR point cloud pipeline	RANSAC ground removal, DBSCAN clustering
sensor_fusion	Multi-sensor fusion	Extended Kalman Filter, camera-LiDAR projection
object_tracking	Multi-object tracking	ByteTrack, per-track Kalman filters
decision_making	Behavioral decisions	Finite State Machine (5 states)
path_planning	Trajectory generation	Hybrid A*, cubic spline smoothing
vehicle_control	Throttle/steering	PID (longitudinal) + Pure Pursuit (lateral)
carla_ad_launch	System orchestration	Sequenced launch files

Quick Start

Prerequisites

• Ubuntu 24.04
• NVIDIA GPU with CUDA support
• Docker with nvidia-container-toolkit

1. Setup Environment

cd Project_1
chmod +x scripts/*.sh
bash scripts/setup_environment.sh
source ~/.bashrc

2. Start CARLA Simulator

bash scripts/launch_carla.sh

3. Build Workspace

cd carla_ad_ws
colcon build --symlink-install
source install/setup.bash

4. Launch Full Stack

ros2 launch carla_ad_launch full_stack.launch.py

5. Monitor (in separate terminals)

# View all topics
ros2 topic list

# Monitor detections
ros2 topic echo /tracking/tracked_objects

# Monitor decisions
ros2 topic echo /planning/driving_command

# Check rates
ros2 topic hz /carla/camera/image

ROS2 Topic Map

/carla/camera/image          ─→  Perception Node
/carla/lidar/points          ─→  LiDAR Processing
/carla/gnss                  ─→  Sensor Fusion
/carla/imu                   ─→  Sensor Fusion
/carla/ego/odometry          ─→  Planner, Controller
/carla/traffic_lights        ─→  Decision Maker
/perception/detections_2d    ─→  Sensor Fusion
/perception/lidar_objects    ─→  Sensor Fusion
/perception/fused_objects    ─→  Object Tracker
/tracking/tracked_objects    ─→  Decision Maker
/planning/driving_command    ─→  Path Planner
/planning/trajectory         ─→  Vehicle Controller
/control/vehicle_cmd         ─→  CARLA Bridge

Decision Making States

State	Trigger	Behavior
LANE_FOLLOWING	Default	Follow lane at target speed
STOP_AT_LIGHT	Red/yellow traffic light	Smooth deceleration to stop
LANE_CHANGE	Slow vehicle ahead	Execute lane change maneuver
OBSTACLE_AVOIDANCE	Static obstacle	Generate avoidance trajectory
EMERGENCY_STOP	TTC < 2 seconds	Maximum braking

Project Structure

Project_1/
├── README.md
├── scripts/                    # Setup & launch scripts
├── docker/                     # CARLA Docker config
└── carla_ad_ws/               # ROS2 workspace
    └── src/
        ├── carla_ad_msgs/     # Custom messages
        ├── carla_bridge/      # CARLA interface
        ├── perception/        # YOLOv8 detection
        ├── lidar_processing/  # Point cloud processing
        ├── sensor_fusion/     # EKF fusion
        ├── object_tracking/   # ByteTrack tracker
        ├── decision_making/   # FSM decisions
        ├── path_planning/     # Hybrid A* planning
        ├── vehicle_control/   # PID + Pure Pursuit
        └── carla_ad_launch/   # Launch files

Configuration

All nodes are configurable via YAML parameters in each package's config/ directory. Key parameters can be adjusted at launch time.

License

MIT License