Device Uplifting and Navigational Group
During the 2024–2025 academic year at Boston University, our senior design team, D.U.N.G. (Device Uplifting and Navigation Group), set out to tackle a challenge in general aviation: the difficulty of maneuvering aircraft within hangars. Traditional aircraft handling often requires several people, precise coordination, and a good deal of patience. Small misjudgments can lead to “hangar rash,” a term pilots use for damage that might look superficial but can be structural, and absolutely becomes costly. Our goal was to design a fully autonomous aircraft maneuvering system that could reposition planes safely and efficiently without human intervention. This team was made up of 6 people, Yousif Al-Hajji, Lorenzo Barale, Zaid Bhatti, Arnav Singh, Zachary Wu, and me, Maysarah Sukkar. An image of our final fleet of robots is below.
Identifying the Problem
The project began with interviews and field visits to Mansfield Municipal Airport, where we spoke directly with pilots about their experiences moving planes in and out of hangars. Every pilot emphasized the same core pain points: avoiding collisions, minimizing physical effort, and maintaining precise alignment within tight spaces. Some used manual tow bars, others motorized tugs, but all agreed that these solutions still required constant attention and help from others. Below are some images of existing solutions, which still require a good deal of effort to operate.
After analyzing these insights, our team defined a clear objective: build a multi-robot system that could lift, coordinate, and navigate aircraft autonomously in cluttered hangar environments — safely, repeatably, and without human guidance.
Research and Early Inspiration
To inform our design, we benchmarked a range of existing products — from handheld electric tugs like the Aircraft Caddy 4K to larger remote-controlled systems such as the AC TrackTech T1V2. While these systems reduce physical strain, they still rely heavily on user input and lack true autonomy.
A pivotal moment in our concept development came from a visit to the Amazon Robotics facility in North Reading, MA. There, we learned about swarm coordination and autonomous mobile robot design directly from Amazon engineers. The visit inspired many of our later design choices — particularly our use of fiducial markers (April Tags) for navigation and a star network topology for robot communication.
Concept Development and Downselection
The project was divided into several core subsystems: mobility, lifting and attachment, sensing, embedded systems, and software/UI. Each subsystem underwent a structured downselection process using pairwise comparison matrices, in which options were evaluated on feasibility, cost, and reliability.
Mobility: We selected a holonomic caster configuration with two motorized and two passive casters arranged in an “X” formation. This layout provided the maneuverability needed to operate in confined spaces while maintaining simplicity and modularity.
Lifting System: We chose a ramp/cradle design to secure and lift the aircraft’s front wheel. The ramp ensured compatibility across different tire sizes and reduced the need for custom parts.
Sensing: We integrated an IMU, encoders, and AprilTag vision system, balancing accuracy with affordability.
Control and Communication: Each robot used an ESP32-S3 microcontroller and a Raspberry Pi 4 for higher-level vision processing, communicating via the ESP-NOW protocol for low-latency synchronization.
User Interface: We built a web application using the MERN stack (MongoDB, Express, React, Node.js), enabling pilots and maintenance crews to schedule movements, track aircraft, and manage maintenance remotely.
Building the System
Mechanically, the chassis was built from aluminum extrusions for modularity and strength, while subsystems were 3D printed in PLA for rapid iteration. This allowed us to refine components like the caster assemblies and cradles within days rather than weeks. Each design iteration revealed new challenges: from caster “wiggle” due to bearing tolerances to friction issues that affected control accuracy. Below is an image of the final motorized caster wheel assembly.
On the software side, we constructed a complete motion pipeline based on AprilTag mapping. Each tag in the environment represented a waypoint, and an adapted Floyd-Warshall path-planning algorithm calculated the optimal route through the hangar. This approach provided reliable navigation while maintaining real-time responsiveness. A map of how the April tags are connected is below.
The embedded system linked everything together: the Raspberry Pi streamed video from the onboard camera, processed fiducials, and relayed commands to the ESP32, which handled precise motor control. Each robot could operate individually or as part of a synchronized team. A schematic of the PCBs we designed for each robot is below, showing the connections between the boards.
Additionally, an image of our final robot design is below.
Testing and Iteration
Once our first prototype was assembled, testing began in earnest at BU’s RASTIC facility. Early trials revealed drift and overshoot during line-following due to uneven wheel friction, so we implemented a PID controller to stabilize performance across all wheels.
Camera testing was another major milestone. At first, motion blur made AprilTags unreadable when the robot was moving. We solved this by adding a ring light, increasing shutter speed, and tilting the camera mount for forward-looking vision — a setup that significantly improved detection reliability. We also added green “slow zones” before fiducials to allow the robot time to read each tag accurately. An image of how the Raspberry Pi and the camera are attached to the robot is below.
In parallel, the web interface underwent extensive testing and refinement. We improved the user dashboard, added scheduling filters, implemented a digital emergency stop, and allowed maintainers to assign “maintenance” or “ready” statuses to each aircraft. The app provided a simple but powerful control center for managing the robotic fleet.
Final System and Results
By the end of the project, our team had successfully built three working robots capable of lifting and maneuvering a scaled aircraft model through a mapped hangar layout. The final system could:
Lift and stabilize a small aircraft wheel using the cradle system
Navigate autonomously between waypoints using AprilTags
Communicate between robots and a central computer via ESP-NOW
Allow users to schedule, monitor, and control movements through the web interface
The project concluded with live demonstrations, showing the robots successfully coordinating to move a mock airplane between parking, hangar, and maintenance areas. A video presentation of the system in its entirety is below.
Lessons Learned and Future Work
Developing D.U.N.G. was a true systems-level engineering experience. We learned to integrate mechanical, electrical, and software design under real-world constraints — from tuning PID loops to debugging communication between boards.
If continued, the next stages for this project would include sensor fusion for 3D localization, full-scale mechanical testing, and integration with real aircraft tires and surfaces. With these improvements, the system could evolve from a proof-of-concept into a scalable commercial product for general aviation.
Acknowledgments
We would like to thank Professor Kenneth Sebesta for his guidance as client and advisor, Professor J. Scott Bunch for his continued support throughout the year, and Gabriel Hebert from Amazon Robotics for his invaluable mentorship during the design process. Most importantly, we’d like to recognize the pilots at Mansfield Municipal Airport for helping us understand the real-world challenges this project faces.
Comments