AERO 465 / Senior aerospace lab / 2025
An autonomous target-detection and engagement demonstrator combining ultrasonic sensing, encoder feedback, embedded control, custom mechanical hardware, and a physically actuated Nerf blaster.
01 / Project brief
GORDON started as our final project for AERO 465. The requirements were open-ended: integrate at least four devices into an aerospace-related system, finish within a month, and stay inside a $25 project budget.
We decided to make things a little harder for ourselves. Our idea was an autonomous Nerf turret that could scan its environment, detect a target, determine its angle, physically rotate toward it, verify its orientation using encoder feedback, and fire without manual input. We had four weeks to make it work.
02 / My role
My role sat heavily at the interfaces between mechanical hardware, electronics, and software.
Team attribution: Timothy Lunney primarily developed the ultrasonic object-detection code. Michael Beath primarily developed the platform-control code and worked with me on the orientation subsystem and rotating platform. I owned most final mechanical and electrical integration, the firing subsystem, software integration, and test-matrix creation.
03 / System architecture
The final system combined three major subsystems: detection, orientation, and firing. Individually, they could work. Integration was where their assumptions collided.
04 / Mechanical design
I designed essentially all final 3D-printed components in SolidWorks: the sensor mount, blaster mounting hardware, geared-motor integration, trigger hardware, custom gears, and the smaller parts needed to make the assembly physically coexist.
The rotating platform was co-designed with Michael and used a lazy-Susan bearing to simplify manufacturing while giving us a smooth rotational interface. The gearing had to fit the available geometry, interface with the encoder, and reduce the geared motor speed to something useful for target alignment.
We changed to straight-cut gears because they made integration and rapid prototyping much more direct. The target reduction could be set by choosing an integer tooth-count pair, then translated into a predictable gear diameter and center distance while we revised the motor location and platform geometry. That made it quick to print a new pair, test the mesh, adjust the ratio if needed, and reinstall the hardware. Compared with the helical concept, the straight-cut geometry also avoided axial loading and removed helix alignment as another variable during assembly. For this prototype, that simpler relationship between tooth count, ratio, and packaging was more valuable than the smoother, quieter mesh of helical gears.
05 / Detection
The detection system used an HC-SR04 ultrasonic sensor mounted on an SG90 servo.
The servo swept through 180° while distance measurements were collected. I implemented the 1D Kalman filter used to stabilize the ultrasonic measurements and integrated it into the final detection logic. When an object entered the 80 cm tracking range, the system stored the corresponding servo angle and passed it to orientation.
06 / Orientation
A 12 V geared DC motor rotated the platform while a rotary encoder measured angular position.
The control logic was intentionally simple: determine orientation error, command the motor in the appropriate direction, and stop once the error entered an acceptable deadband. This was closed-loop positioning with encoder feedback, not PID control.
Michael primarily developed the platform-control code. I helped establish requirements, design the mechanical integration, debug closed-loop behavior, and bring the orientation subsystem into the complete system.
07 / Firing system
This was my subsystem.
I owned most firing-system development and integration. The logic first spun up the Nerf blaster flywheels; after a delay for them to reach operating speed, the trigger actuator fed the dart into the flywheels. I wrote the firing-sequence code and integrated the engagement conditions, flywheel control, actuator timing, and physical trigger mechanism.
The code was the straightforward part. The mechanical trigger was where things became interesting.
08 / Engineering under constraints
The messy parts were not side stories. They were the engineering story.
The linear actuator we wanted was around $40. Our entire project budget was $25. I started looking for anything else that could provide the same short linear motion and landed on an automotive door-lock actuator. Cost: about $3.
The door-lock actuator worked, then its startup current spike killed our original 12 V supply. We needed 12 V with much more current capability. The practical answer was a car battery, plus a fuse. It powered final testing and demonstration, after temporarily rendering our transportation unusable.
The actuator was fast enough to move the trigger before the dart could feed correctly into the flywheels. The final fix was zip ties with carefully tuned slack. That small mechanical delay gave the dart time to feed and made the firing sequence reliable.
09 / System integration
Detection, orientation, and firing were developed separately. The final demonstration required all three to agree on what happened next.
I performed the overall software integration while also integrating the physical hardware and electrical system those software blocks commanded. Sometimes a software problem was wiring. Sometimes a mechanical problem was timing. Sometimes the component itself was bad.
10 / Testing
I created the project test matrix to evaluate combined targeting performance.
A laser pointer mounted parallel to the blaster barrel provided a consistent targeting reference. For each trial, the target was placed at a predetermined location, GORDON detected and aligned itself, and we measured deviation from the intended target center.
The measurement included the combined effects of sensing, target-angle determination, gearing, platform motion, encoder feedback, mechanical assembly, and firing alignment.
11 / Results
>7 in
Early trials remained relatively close to target center. As testing continued, positioning error became progressively larger; later trials exceeded approximately seven inches of deviation.
The prototype was not fully resetting or recalibrating position between acquisitions, so small errors accumulated across sequential trials. The four-week schedule did not leave time for a larger repeated-trial dataset. These results validate the prototype while clearly identifying the next engineering problem.
12 / Next iteration
13 / Personal takeaway
What stayed with me was not simply getting an autonomous Nerf turret to work. It was how quickly the engineering changed every time we left the ideal case: the actuator cost too much, the replacement actuator killed the supply, the new source fixed that, then the actuator created a timing problem. The gear design changed. MOSFETs burned out. A faulty breadboard sent us looking in the wrong direction.
Sometimes the solution was software. Sometimes it was a redesigned part. And sometimes it really was a car battery and a few strategically placed zip ties.
14 / Project archive
Select any photograph to inspect its full framing.