Mesosphere

Deployment

Mesosphere is my Cal Poly Computer Engineering senior capstone (Capstone I and II, January to June 2025), built by a team of four for OWL Integrations, the company behind Project OWL and the ClusterDuck Protocol, with an EE faculty client on campus. It is a completed field-test program with a year-end report in Cal Poly’s Digital Commons, not a hosted product. The code went to a private repository and the hardware passed to the next student team in fall 2025. This is a different project from the Sentinel Protocol, which was my senior project the following year.

Problem

High-altitude and field telemetry has to keep working where connectivity is unreliable, and it needs camera-based situational awareness on top of the raw sensor feed. Both of those cost power. The platform had to do more, for longer, on a battery budget that does not grow just because the mission gets harder.

Build

The brief from the client was to virtualize the ClusterDuck Protocol, the mesh radio firmware OWL runs on its “Duck” devices, so the same mesh could run on Raspberry Pi hardware. We built the mesh on Raspberry Pi Zero 2 W boards with LoRa hats, added boards through the spring to run real mesh experiments, and paired it with a YOLOv12 edge camera for on-board object recognition and event-triggered notifications. My job was the adaptive power-management strategy: an ML-driven controller that traded compute and transmission cadence for battery life based on what the mission needed at each moment rather than a fixed duty cycle. I also owned the schedule (the Gantt chart and weekly agenda the program manager reviewed), worked the LoRaWAN and ClusterDuck integration alongside the camera pipeline, and built the Power BI dashboards the team used to review flight and energy data after each field test.

Presenting was also my job: the design-build-test milestones, the client review calls, the Spring EXPO demo, and the results in front of USAF and xTechSearch stakeholders.

Proof

Video

In field tests, the adaptive power-management strategy improved mission duration by about 25% and cut power consumption by about 30%. Those two numbers came from the same set of field trials, and they are the reason the platform could stay airborne longer on the same battery instead of trading flight time for camera uptime.

After the Spring EXPO the program manager passed on the client’s verdict in writing: they were pleased with the demo and the progress. The Computer Engineering department nominated the team for a student project spotlight, and the faculty client invited us to demonstrate the LoRa mesh at the IEEE International Microwave Symposium amateur radio reception at Moscone Center in June 2025.

What I would do next

I would instrument the adaptive controller with finer-grained telemetry so the power-versus-duration tradeoff could be tuned per mission profile instead of using one strategy for every flight. I would also push the YOLOv12 pipeline further at the edge, so object-recognition events themselves could trigger the power controller instead of running on a separate schedule. Field photos and diagrams from the OWL Integrations test runs would make this page stronger than the current concept visuals, and I would go get them before the next update.