Coral Reef Restoration Robot

Inspiration

After a visit to the Georgia Aquarium, I was inspired to do something to help preserve the ocean's ecosystems. Pollutants cause algal blooms which starve coral reefs of oxygen and sunlight. Inspired by the news of an underwater robot being deployed to kill thorned starfish in coral reefs, I decided to make one that removes algae from coral reefs.

What it does

This robot uses computer vision, depth camera, and lidar readings to accurately detect the location of algae patches. After approaching the algae, it sprays it flucanozole to kill it and vacuums up the remaining algae. It also uses activated carbon to filter out nitrates and phosphates from the water to further slow down algal growth.

How I built it

I designed and assembled the robot parts in Solidworks to produce an exploded view. In addition to the CAD, I annotated images from the internet using CVAT, generated the dataset using Roboflow, trained the YOLOv5 model using Google Colab to obtain the weights file, and tested the model using the Ultralytics official YOLOv5 GitHub repo on a video from a diver's perspective I found on the internet.

Challenges I ran into

It was difficult to find images of algae in coral reefs where the algae was distinct enough from the environment such that I could annotate it. Finding 28 images took 3+ hours. It was also difficult to CAD multiple parts and visualize how they would interact with each other in the assembly while still looking somewhat realistic. Multiple parts had to be redesigned.

Accomplishments that I am proud of

I am proud to have assembled the full robot in the short time that I had. I am also proud of having been able to research the issue of algae removal and ideate a robot that proposes multiple solutions.

What I learned

I learned a new technique to CAD realistic propellors in Solidworks using curves projected onto custom twisted surfaces. I also learned how to efficienty train and deploy a YOLOv5 computer vision model.

What's next for Coral Reef Restoration Robot

The parts need to be 3D-printed and further redesigned. A microcontroller needs to be programmed and wired to the lidar, depth camera, lights, vacuum pump, motor, propellors, front cam, and spray tank. The CV model needs to be improved to provide higher accuracy results. Flow simulations also need to be performed to judge how hydrodynamic the robot actually is. After that, experimental trials can begin.

Built With

• colab
• cvat
• python
• roboflow
• solidworks
• yolov5