Learning Trajectory Tracking For An Autonomous Surface Vehicle In Urban Waterways

Roboat is an autonomous surface vessel (ASV) for urban waterways, developed as a research project by AMS Institute and MIT. The platform can provide numerous functions to a city, such as transport, dynamic infrastructure, and an autonomous waste management system.The goal of this project is to develop a learning-based controller for the Roboat platform with the goal of improving robustness and generalization. When subject to uncertainty in the model or external disturbances, the proposed controller should be able to track set trajectories with less tracking error than the current Nonlinear Model Predictive Controller (NMPC) used on the ASV. To achieve this, a simulation of the system dynamics was developed based on research done on the platform and previous literature. The simulation also includes the modelling of the necessary uncertainties and disturbances. In this simulation, a trajectory tracking agent was trained using the Proximal Policy Optimization (PPO) algorithm which was then validated and compared to the current control strategy both in simulation and in the real world.