Autonomous Reconnaissance Drone

3D Printing Integration Fusion 360 Python

I worked on a drone platform used to demonstrate autonomy, vision, and RF sensing software developed by a defense-focused company. My role was focused on mechanical integration and testing. That included mounting cameras, onboard compute, radios, and RF hardware, dealing with wiring and shielding issues, and making sure parts could be removed and replaced without tearing the whole system apart. A lot of the work came down to fit, layout, and what happens when the drone lands harder than planned.

Role

Mechanical Engineering Intern

Goal

Build a flight-ready platform that reliably demonstrates autonomy, vision, and RF sensing in real operating conditions.

Overview

The platform combined cameras, onboard compute, radios, and RF detection into a single airframe. From a mechanical standpoint, the challenge was keeping everything accessible and protected while testing was ongoing. Once the full RF stack was powered, layout and shielding started to matter for GPS performance, so parts were moved and rerouted as issues showed up. The goal was a setup that could be flown, adjusted, and repaired without turning every change into a rebuild.

Figure 1 — Ground configuration during integration and testing — **Figure 1 — System configuration.** Ground reference view used during integration and testing. (Click image to expand.)

Design Approach

Integration decisions were driven by what was most likely to break first. Impacts, vibration, cable fatigue, and RF interference all showed up during testing. The priority was keeping expensive hardware protected and easy to access, even if that meant giving up some packaging efficiency. RF layout also became important once GPS dropouts started appearing with certain hardware combinations.

Constraints & considerations

Packaging: fit sensors and compute without blocking access or airflow.
Serviceability: parts that break or get handled often needed fast removal.
RF effects: layout and shielding affected GPS lock and stability.

Designed Failure: Landing Gear Breakaway

After a few crashes during early testing, it became clear that the landing gear should take the damage instead of the frame or electronics. I designed the landing gear mount with a defined weak point so it would crack in one predictable location. The geometry was adjusted in Fusion 360 using an impact-style load case and then tested through actual crashes. When it failed, it failed where intended, which kept repairs simple.

Issue: rigid landing gear transfers impact loads into the frame.
Change: introduce a controlled weak section in the mount.
Result: repeated failures at the same spot, protecting the rest of the drone.

Figure 2 — Breakaway landing gear mount with controlled failure location — **Figure 2 — Breakaway mount.** Geometry tuned so the mount splits at a known location. (Click image to expand.)

Supporting Fabrication

Figure 3 — 3D printed mounts and housings used during integration — **Figure 3 — Rapid fabrication.** Printed parts used to support iteration and replacement. (Click image to expand.)

3D printing was the main way changes were made quickly. I designed and printed battery cases, sensor mounts, adapters, and protective housings so components could be moved or replaced without reworking the entire layout. These parts were designed around fit, access, and replacement rather than appearance, and were iterated directly from test feedback.

Vision Detection (OpenCV)

Figure 4 — OpenCV-based detection output used during testing — **Figure 4 — Detection output.** OpenCV-based pipeline used during detection experiments. (Click image to expand.)

I built a Python and OpenCV pipeline to support detection when drones appeared very small against the sky. The approach focused on finding the horizon to reduce the search area, then flagging motion-based candidates. This ran alongside a YOLO-based detector: when YOLO could classify a target, it took priority, and when it could not, the OpenCV pipeline still provided usable information for the operator and downstream systems.

What this enabled

Fallback detection: visibility when learned models struggled.
Hybrid approach: OpenCV complemented YOLO instead of replacing it.
Tuning support: overlays made it easier to see failure cases.

Results

Integrated cameras, RF hardware, and onboard compute into a compact drone platform.
Designed a breakaway landing gear mount that reduced damage during crashes.
Used 3D printing to support fast mechanical iteration and repair.
Developed an OpenCV pipeline that complemented YOLO-based detection.

Figure 5 — Integrated drone configuration used during testing — **Figure 5 — Integrated platform.** Configuration used during testing and iteration. (Click image to expand.)

What I’d Change Next Time

Address RF layout and shielding earlier to avoid GPS issues later.
Simplify wiring paths further to reduce maintenance time.
Reduce overall size while keeping the same level of modularity.
Turn the vision prototype into a more repeatable test module.

← Back to projects Back to top ↑