Hluboké učení pro autonomní off-road řízení v simulaci

Abstract

This thesis presents different ways to make a car autonomous. We will use the power of machine learning and neural network to ?teach? a car how to drive autonomously in an off-road environment by using only a minimum set of sensors, in our case which is just a single RGB camera. We will first focus on a technique called imitation learning, it is a supervised learning algorithm which takes a lot of example pairs (image; driving command) to extract a policy that the car will use to drive in unseen situations. Then we will use the so-called reinforcement learning technique. It is an unsupervised learning algorithm which manages, by a lot of trial and error experiments, to create a policy used by the car to drive safely. We managed with these two techniques to make our car drive itself in a simulator.

This thesis presents different ways to make a car autonomous. We will use the power of machine learning and neural network to ?teach? a car how to drive autonomously in an off-road environment by using only a minimum set of sensors, in our case which is just a single RGB camera. We will first focus on a technique called imitation learning, it is a supervised learning algorithm which takes a lot of example pairs (image; driving command) to extract a policy that the car will use to drive in unseen situations. Then we will use the so-called reinforcement learning technique. It is an unsupervised learning algorithm which manages, by a lot of trial and error experiments, to create a policy used by the car to drive safely. We managed with these two techniques to make our car drive itself in a simulator.