Autonomous Blimp
Graduation research for the Maintenance Lab (HvA): a real-time navigation system for an indoor blimp with a maximum payload of 850 grams. The goal was autonomous flight in large indoor spaces such as the atrium without an external computer, fully onboard on a Raspberry Pi 5 with ROS 2. I designed and built a modular system for perception, planning, and control. Field tests with the first design (V1) revealed where things went wrong: a stereo baseline that was too short for large spaces, and steering lock-up caused by overly aggressive PID corrections. Based on that analysis, I developed an improved V2 design with a wider camera setup, Pure Pursuit, Control Allocation, and burst & steer.
Role
Software Engineering, Embedded Systems, ROS 2
Sector
Robotics & Predictive Maintenance
Year
2025-2026
V1: test, measure, fail
The first system combined an Intel RealSense D456, RTAB-Map for V-SLAM, RRT* path planning via OMPL, and a reactive PID controller with three speed modes. On paper it made sense; in the atrium it did not.
Three test flights, zero successful missions but valuable data nonetheless. Position estimation jumped due to sensor noise (the 9.5 cm baseline was too short for long-range measurements). The blimp overshot waypoints because of added mass, tried to fly back, and got stuck in correction loops. At the same time, the altitude servos stayed at 100%, making horizontal steering impossible: steering lock-up.
That analysis became the foundation for V2 not guesswork, but measurable failure mechanisms.
V2: software that respects physics
V2 addresses the core problems directly. A modular OAK-FFC stereo camera with a 20 cm baseline theoretically halves depth error. The control logic was rewritten: Pure Pursuit looks ahead instead of correcting backward, Pulse & Coast uses inertia instead of fighting it, and Control Allocation prevents altitude and steering from blocking each other.
Due to delivery delays, V2 could not be physically tested, but the calculations and architecture show that autonomous navigation within 850 grams is feasible as long as sensor, compute, and control are aligned.
Stack: ROS 2, Raspberry Pi 5, RTAB-Map, OMPL, Python/C++, custom telemetry logging.
More projects
Get in touch
I'm always open to hearing from new people about their ideas, projects, or potential collaborations.