Learning and Planning for Autonomous driving

This project, funded by Hyundai Mobis, was initiated to develop a driver model for a rear-wheel steering vehicle to replace skilled drivers in vehicle testing process. The goal was to enhance driving abilities with a hierarchial control system using model predictive control (MPC) and reinforcement learning. Our approach utilized MPC as a low-level policy, leveraging its capacity to optimize systems with complex non-linear dynamics and constraints. Deep reinforcement learning was empolyed as a high-level policy to optimally schedule the parameterized cost function of the MPC. Our method is visualized throught IPG CarMaker with connection of MATLAB/Simulink for Python ML scripts.

*Please note that this is a on-going project and the video on the left is not the result, but just the video I got in mid-progress!

This project, conducted as part of the graduate course 'Optimal Control and Reinforcement Learning (ME6505)', involved the development of a nonlinear Model Predictive Control (MPC). I constructed Bezier curves for position and curve parameters and used it as a reference to track with the MPC. Despite being an undergraduate student, my approach resulted in the fastest lap time among the graduate student participants. 

Paper/publication:
You can see my report of this work under 'files' of the Github repository.

Github repository:
https://github.com/n00Nspr1ng/Nonlinear_MPC_with_Bezier_curve