Singapore is heavily reliant on our maritime activities; it is critical that we keep our waters secure. We aim to use reinforcement learning to create a model that allows autonomous underwater vehicles (AUVs) to collaborate and efficiently search and patrol a pre-determined area. This project was done by Wee Yen Zhe (randomwish) and Wee Yu Shuen (chlwys) as our final year project under the Defense Science and Technology Association (DSTA) and Ngee Ann Polytechnic (NP). Special thanks to our mentors, Mr. Poh Chun Siong (DSTA) and Mr. Lim Ching Yang (NP).
This project uses Unity's ML-Agents package. A realistic environment is created in Unity and agents are trained using Proximal Policy Optimization (PPO). The video below demonstrates the training results over time.
TrainingMovements.mov
For more details, do read our report.