UAV–UGV Coordination in CARLA

A collaborative Florida Atlantic University senior-design project exploring how an aerial follower can track a ground vehicle inside CARLA. Modular controllers exchange position and waypoint messages, while a coordination state machine handles tracking, stale telemetry, search, and mission completion.

Implemented capabilities

• Ground vehicle navigation through CARLA autopilot or scripted routes using BasicAgent when CARLA's agent modules are available.
• A kinematic aerial follower with speed, acceleration, vertical-motion, and yaw-rate limits, plus an attached RGB camera.
• Follow waypoints calculated from ground-vehicle heading, distance, altitude, lateral offset, and velocity lookahead.
• Target-loss detection based on stale ground-vehicle position messages, with hover and spiral search behavior.
• LiDAR processing and obstacle telemetry, an operator GUI, and timestamped CSV logging of positions, obstacles, and state changes.

Architecture

flowchart LR
  CARLA[CARLA world] --> UGV[UGV controller]
  UGV -->|position and obstacles| Broker[In-process message broker]
  Broker --> Coord[Coordination state machine]
  Coord -->|follow waypoints| Broker
  Broker --> UAV[Kinematic UAV controller]
  UAV -->|position| Broker
  Broker --> Logger[CSV data logger]
  Broker --> GUI[Operator console]
  UAV --> CARLA

Loading

The aerial vehicle is a CARLA actor or sensor platform moved by a kinematic model, since CARLA does not provide native UAV flight dynamics. Tracking uses vehicle telemetry; the RGB feed does not establish computer-vision target tracking.

Run on the documented Windows setup

Prerequisites: Windows 10/11, Python 3.12, a CARLA 0.9.16 server, and its matching Python API. Availability of the CARLA package depends on platform and Python version.

git clone https://github.com/OmElMon/G27-Project.git
cd G27-Project
py -3.12 -m venv .venv
.venv\Scripts\Activate.ps1
pip install -r requirements.txt

If the pinned CARLA package cannot be installed, use the compatible API distribution from your CARLA installation. For scripted routes, make CARLA/PythonAPI/carla available on PYTHONPATH so agents.navigation imports successfully. The code also checks a project-local Carla/PythonAPI/carla directory.

Launch the CARLA server separately, wait for its map to load, then run:

python main_gui.py
# Or use the terminal runner:
python main.py

The runners prompt for navigation mode, follow distance, altitude, and camera target. Stop with the GUI Quit button or Ctrl+C. Logging uses a timestamped directory under the configured log directory, defaulting to logs/.

Source guide

File	Role
main.py, main_gui.py	Simulation orchestration and lifecycle
ugv_controller.py	Ground navigation and telemetry
uav_controller.py	Aerial follower and kinematic movement
coordination_platform.py	Follow geometry and state transitions
message_broker.py	Publish/subscribe communication
gui_console.py	Operator interface
data_logger.py	CSV recording and separation metrics
config.py	Runtime parameters and topics
ed2_avoid.py, sensor_manager.py	Additional obstacle/sensor code
reference/	Earlier reference implementation

Validation and limits

Subsystem entry points include python coordination_platform.py for simulated-message coordination exercises and CARLA-dependent python ugv_controller.py / python uav_controller.py. These are manual exercises, not an automated assertion-based test suite. The repository has no GitHub Actions runs at the time of this audit. A CARLA simulation was not executed for this documentation update.

The terminal runner parses --host and --port, but its connection method currently uses configuration constants rather than those parsed values; configure the connection in config.py until that wiring is fixed. Reproducible scenario tests, quantitative following-error plots, and improved sensor fusion are useful next steps.

Team and attribution

This is a team project; module ownership annotations in source should be preserved.

• Evan Frisone — Team Leader, Computer Science
• Omar Elharbili — Computer Science
• Sean Bowden — Computer Science
• Roberson Robert — Team Member
• Syeda Haque — Computer Science

Sponsor / mentor: Dr. Xiangnan Zhong, Florida Atlantic University.

No license file is currently included; no open-source license is asserted here.