A project exploring autonomous surface vessel navigation and collision avoidance using reinforcement learning.
The environment is a custom Gymnasium environment that simulates the ASV navigation task. This environment simulates vessel movements using an action space that controls acceleration and rudder position.
Train contains code that trains a Soft Actor-Critic model to learn how to navigate to the target quickly and avoid obstacles.
Testing simulates the model on a new randomly initialized environment based on the model file from specific weights and biases run-id.