Check the website here

AeroFusion

Summary

AeroFusion develops a sensor-driven framework that enables a biologically inspired BLIMP UAV to detect, track, and approach targets autonomously in dynamic indoor environments. The system fuses IMU, barometer, and camera data within a modular ROS 2 stack and provides real-time telemetry through a browser-based GUI.

Highlights

• Perception: Custom YOLOv5 model with quadrant logic for target tracking.
• Control: PI and differential drive controls; force allocation via pseudo-inverse.
• Sensor Fusion: IMU + barometer filtering for stabilized altitude and attitude.
• ROS 2 Stack: C++ and Python nodes with explicit topic interfaces and launch files.
• GUI: WebSocket (rosbridge + roslibjs) interface with live video and telemetry.
• Validation: Thrust-stand characterization and Vicon-based motion tracking.

Media

CAD rendering of the BLIMP platform.

Browser-based ROS telemetry and controls.

Architecture

/image_raw → YOLOv5_Detector → /blimp/detected_position
/imu_data + /barometer_data + /balloon_input → InvKine → /forces
/forces → F_to_ESC → /ESC_balloon_input
/ESC_Manual_input + /ESC_balloon_input → ModeSwitch → /ESC_input → ESC_Driver

Key nodes:

• detect_cpp (C++): camera detection and goal/balloon modes.
• inv_kine (C++): force estimation from IMU/barometer/vision.
• f_to_esc (C++/Eigen): body forces → ESC PWM via pseudo-inverse.
• esc_driver (Python/pigpio): PWM output and ESC arming.
• joy_to_esc (Python): manual teleop mapping.
• mode_switch (Python): manual vs autonomous arbitration.
• read_imu, read_altitude (Python): sensor interfaces.

Modeling and Identification

• 6-DOF dynamics: rigid + added mass, Coriolis/centripetal, linear damping, restoring forces.
• Allocation: non-square thruster map solved with Moore–Penrose pseudo-inverse.
• System ID: SolidWorks inertia, Lamb k-factors for added mass, thrust-stand linear PWM→thrust fits, and damping via flight tests with Vicon.

Representative figures:

• Thrust vs PWM: Figure 4/5
• Block diagram: System diagram

Implementation Details

Platform: Raspberry Pi, ROS 2 Humble, rclpy/rclcpp, Eigen, pigpio, OpenCV, roslibjs, rosbridge_server, web_video_server.

Selected topics:

• Perception: /image_raw, /blimp/detected_position
• Sensing: /imu_data, /barometer_data
• Control: /forces, /ESC_balloon_input, /ESC_Manual_input, /ESC_input

How to Run (excerpt)

# Workspace
mkdir -p ~/blimp_ws/src
cd ~/blimp_ws/src && git clone https://github.com/RAS598-2025-S-Team03/BLIMP-Packages.git

# Build interfaces first, then all
cd ~/blimp_ws && colcon build --packages-select blimp_interfaces
source install/setup.bash
colcon build

# Optional: Oak-D Lite camera stack in separate workspace
# (see project docs for install script and rosdep)

# Verify joystick
ros2 run joy game_controller_node

# Arm ESCs
ros2 launch auto_control arming.py

# Launch full stack
ros2 launch auto_control updated_launch.py

# GUI bridge + camera stream
ros2 launch blimp_gui rosbridge_camera_launch.py