Daniel Kim's Autonomous Underwater Glider Aims to Seek Out Sub-Sea Trash

High school junior Daniel Kim is creating an autonomous submarine designed to seek out underwater trash, and has chosen to release everything — from CAD files to the source code — under a permissive open source license.

"This project aims at creating an autonomous underwater vehicle (AUV) that utilizes cameras, sensors, a microcontroller, and GPS to categorize, identify, and map underwater trash," Kim explains of the OceanAI effort. "Detection of litter will be performed with a deep learning model — optimized to run on a microcontroller — allowing for cheap and power-efficient inference of underwater footage."

OceanAI is a work-in-progress, but already in an impressive state: The control board, software, and vehicle design have all been completed, complete with a "Command Center' user interface that provides readouts of system status and hardware health.

Kim began the project by drawing on his experience with model rocketry, and there's little surprise then that the submarine, which results bears quite some resemblance to a rocket — shaped, as it is, like a tube with a rounded nose-cone but boasting two large "wings" to the sides to allow it to operate as what Kim calls an "underwater glider."

Inside the tube is a ballast tank to allow the submarine to rise and sink, batteries for untethered use, and a custom control board, which includes connectivity to a sensor board with dissolved solids sensor, a pressure sensor, a thermister, and the camera that will eventually provide input to the trash-seeking on-board AI. A floating "transmitter board," meanwhile, is designed to sit outside the submarine on the water's surface in order to provide access to global navigation satellite system (GNSS) radio signals and to transmit to a controlling terminal.

The project's source code and design files are now available on the OceanAI GitHub repository, under the permissive MIT license.