This repository implements an imitation learning pipeline for AUV control. It uses the "FossenNet" neural network to mimic an optimal NL-MPC policy and includes tools for data generation, training, and real-time C++ inference on GPUs.

A modular, GPU-accelerated control system framework for AUV. Features Model Predictive Control (MPC) with CasADi, CUDA-based environment mapping, GPU-accelerated A* path planning, Fossen equations for marine vehicle dynamics, and supports imitation learning by distilling CasADi-based nonlinear MPC models into NN controllers.