Inspiration
There are more than 150 million tonnes of plastic in our oceans. This causes huge damages to ecosystems and generate massive costs linked to beach cleaning, tourism and fisheries profit losses. Recently, large collecting nets were developed (www.theoceancleanup.com) but these need human guidance and intuition in order to efficiently collect litter.
What it does
We decided to automate the collecting step by building autonomous agents that learn from ocean currents and simulations.
How we built it
Reinforcement learning agent trained to automatically and optimally collect plastics in oceans and seas.
Challenges we ran into
Physical modeling of the coast line. Long training time.
Accomplishments that we're proud of
Strong Lagrangian simulator to simulate plastic debris drifts. Well functioning reinforcement learning model accounting for current flows, physics and simulator outputs.
What we learned
Oceans are dirty and reinforement learning works in practice.
What's next for AlphaNet
Refining the model, better memory of events and multi-agent modeling.
Log in or sign up for Devpost to join the conversation.