CORALCAST
Concept only — will not be built
The Idea
An autonomous surface robot for shallow-water coral reef health monitoring. The robot floats and roams a shallow reef flat (roughly 2–10 ft deep), using a downward-facing Raspberry Pi Camera Module 3 angled about 10° to photograph the reef every ~2 seconds and transmit each image to a separate processing device (phone or laptop) on shore. That base station runs image recognition to classify coral as healthy or dead/bleached, and logs the GPS coordinate of any flagged frame onto a reef health map — turning a manual snorkel/dive survey into something closer to automated and repeatable, without needing divers, boats, or expensive marine robotics hardware.
What I Explored
Camera & Imaging
- Camera choice — settled on the Raspberry Pi Camera Module 3 for its sensor, over other options considered
- Whether sun glare, waves, and surface distortion would still allow usable downward-facing shots — flagged as something that can only really be confirmed by testing in the water
Communication & Data
- Wireless link design targeting ~1km range at up to 8Mbps transfer, without a settled solution yet
- Keeping all coral-classification intelligence on the base station instead of the robot, so the field unit only needs to capture and transmit
Mechanical & Environmental
- Making the enclosure watertight while still serviceable/openable for repairs
- Robot stability against waves, wind, and currents, and how drift/tilt would affect image quality and positioning
Positioning & Processing
- GPS module accuracy (2–5m) and what that means for pinpointing a specific coral location
- Deliberately excluding a dedicated processing device from the robot itself — any phone, laptop, or tablet capable of running the recognition pipeline can serve as the base station
What I Learned
The design pass was honest about what's still unverified rather than assuming it would work: underwater image quality through glare and wave distortion needs an in-water test to actually confirm, the wireless communication approach (range vs. transfer speed) isn't settled yet, and GPS accuracy sets a real ceiling on how precisely a flagged coral location can be pinned down. Keeping all classification logic on an external processing device (rather than onboard the robot) was a deliberate architecture choice to keep a single unit buildable within budget. A full parts list and cost breakdown were worked out — landing at roughly ₱11,200 against a ₱15,000 target — but no physical build or water testing has happened yet.
Outcome
Discontinued before any build. With a limited budget for the year, committing to parts (₱11,200 of a ₱15,000 target) only made sense if the design's open risks — underwater image quality, an unresolved wireless link, and robot stability in waves and current — were reasonably likely to work out. They were unproven enough that a failed build would have meant spending the budget with nothing usable to show for the year, so I shelved the idea at the concept stage rather than risk the build.