BIRDS-RPM: redesign a satellite without losing the evidence

BIRDS-RPM, also called Kumo, is a multinational 3U CubeSat project based at Kyushu Institute of Technology. It extends the BIRDS engineering heritage while changing parts of the platform. Its experimental onboard computer, Open Space Imaging Lab and Caburei packet-radio payload investigate different capabilities. A shared program history helps a team reuse knowledge, but each changed combination of hardware and software presents a new integration problem.

The project’s status page, reviewed on 20 October 2026, still identifies flight-model development as the current phase, with flight readiness, launch and operations upcoming. Its calendar includes earlier planned development ranges. Reaching the end of a printed range is not proof of launch. Keep the published development evidence separate from an orbital result until a dated operations report establishes what actually happened.

The experimental onboard computer investigates STM32-based processing, card-edge connections and RS485 communication between subsystems. The team reports radiation and performance testing, including current, temperature and memory-use measurements against the earlier PIC-based approach. That comparison asks more than whether the new processor runs faster: a usable replacement must work with the surrounding power, data and mechanical interfaces.

In an invented power check, a payload allocation is 2.0 watts. Its processor consumes 0.7 watts, the sensor 0.6 watts and an interface board 0.3 watts, leaving 0.4 watts of margin. Replacing the interface with a 0.8-watt version raises the total to 2.1 watts and exceeds the allocation by 0.1 watts. A mechanically fitting board can therefore break the system budget. These values illustrate the reasoning, not Kumo’s measured consumption.

A new interface changes the budget

Open Space Imaging Lab links low-cost imaging hardware and onboard classification to satellite-learning opportunities for students in Rwanda. Its published tests include command-driven image capture, storage and telemetry, with a classifier trained for scene categories. Caburei explores amateur packet radio using digipeating and store-and-forward — open definition. Both require a complete path from a ground request through onboard action to understandable returned data.

For a hypothetical imaging test, suppose a classifier calls twenty images clear but independent review finds five obscured by cloud. Fifteen of the twenty selected images are actually clear: a 75 percent success fraction within that selected group. That does not reveal how many clear images the classifier missed. Preserve a separate evaluation set and inspect both kinds of error before claiming that onboard selection improves the delivered observations.