Drone

Autonomous multirotor UAV mission system for airdrop and reconnaissance competitions.

Refactored and integrated from repos: drone | cv_py | cv_cpp By AGENTS(Claude Opus 4.6) which is FOR REFERENCE ONLY.

ROS 2 Humble / C++20 / NVIDIA Jetson / ArduPilot

Quick Start

# Build (Docker)
./scripts/build.sh

# Simulation (Gazebo SITL)
./scripts/sim.sh

# Run on hardware
source install/setup.bash
ros2 launch drone drone.launch.py

Architecture

Single C++ process on NVIDIA Jetson. Camera capture, TensorRT YOLO inference, detection filtering, position control, and mission logic all execute in one ROS 2 node with zero network hops.

Camera -> TensorRT YOLO -> Kalman Filter -> Clustering -> Mission Handlers
                                                              |
                                                     PosControl (PID cascade)
                                                              |
                                                     MAVROS -> ArduPilot

See docs/ARCHITECTURE.md for the full module dependency graph, data flow, coordinate frames, and CMake target layout.

Mission

Defined in docs/PROBLEM.md:

1. Take off autonomously
2. Fly to the drop zone, detect three containers, drop two payloads
3. Fly to the reconnaissance zone, identify ground markers
4. Return and land on the H-marker using visual servo

Project Structure

include/drone/
  utils/        readyaml, timer, keyboard, rotate
  math/         pid, basic_pid, scurve, types, utils
  perception/   camera_model, detection_filter, clustering, detector (TensorRT)
  control/      pos_control, velocity/cascade/trajectory controllers, mavros_commander
  drivers/      node_base, motors, inertial_nav, servo, gimbal
  mission/      drone_node, handlers (takeoff, airdrop, recon, landing), target_tracker
src/            Matching .cpp implementations + main.cpp
test/           22 test suites, 251 tests
config/         6 YAML config files
launch/         ROS 2 launch file
scripts/        build.sh, sim.sh, build_engine.sh
model/          TensorRT engine files (not tracked, see scripts/build_engine.sh)

Build and Test

./scripts/build.sh              # Docker build + test (251 tests)
./scripts/build.sh shell        # Interactive shell in container
colcon test --packages-select drone  # Run tests natively

TensorRT Engine Building

./scripts/build_engine.sh export model/best_circle.pt         # PyTorch -> ONNX
./scripts/build_engine.sh onnx model/best_circle.onnx --fp16  # ONNX -> TensorRT

Hardware

• NVIDIA Jetson Orin/Xavier (onboard compute)
• ArduPilot flight controller (via MAVROS)
• Camera (V4L2/CSI)
• Servo-actuated payload release mechanism

Documentation

• Architecture: module graph, data flow, coordinate frames
• Usage Guide: build, simulation, run, configuration, testing
• Mission Specification: competition rules
• Dependency Analysis: MAVROS topic graph
• Refactoring Roadmap: 8-phase migration plan
• Progress: phase-by-phase completion status

License

MPL-2.0. Copyright (c) 2024-2026 HDU-DXY-Team.