Marine Autonomy Challenge (MAChallenge 2025)

This project is focussed on investigating the use of marine autonomy in detecting, locating and cleaning ocean plastics using a USV in the context of a competition called the MAChallenge. As part of the research and tasks assigned in the challenge, we will investigate reliable methods for autonomous navigation, berthing, obstacle avoidance and target detection.

Thus, the key components of this project are as follows:

➢ Sensor fusion technique: These techniques aim to utilise strengths of various sensors, to overcome issues such as inaccuracy of GPS signals in urban areas or LIDAR's limitations in heavy fog. By intelligently fusing data, the system can produce a more detailed map of the environment.

➢ Deep Reinforcement learning: This will allow an autonomous agent to learn optimal behaviours by interacting with its environment and the use of neural networks can further enhance its capabilities. By implementing reinforcement learning techniques to map out the cost of potential paths, it will also ensure better obstacle avoidance measures and can lead towards a COLGREGs compliant USV by penalising collision prone trajectories. This will be crucial for enabling the system to make real-time decisions in uncertain complex environments such as in water.

➢ Computer vision: To enable the system to recognise and respond to obstacles, we will attempt to use YOLO frameworks for detecting and classifying objects. The goal is to not only detect obstacles but also predicts their movements to improve responses in real-time. This can be attempted by training neural networks to extract useful information from image data and then predict its motion from sequential data to be able to track positions in a dedicated environment as time progresses since dynamic obstacles will be changing in time.

➢ Seamless control manoeuvring: For ensuring seamless autonomous berthing, a control algorithm will need to be paired with a detection algorithm to enable safely docking in its constraint space.