A self-driving system using double deep Q-learning | |
| Author | Siraphop Prasertprasasna |
| Call Number | AIT Thesis no.ISE-21-19 |
| Subject(s) | Image segmentation Optical radar Reinforcement learning Supervised learning (Machine learning) |
| Note | A thesis submitted in partial fulfillment of the requirements for the degree of Master of Engineering in Mechatronics |
| Publisher | Asian Institute of Technology |
| Abstract | In the self-driving car industry, LIDARs and cameras have been used as sensors to drive a car. LIDARs can know the exact distance around a car, but when a car is driven by a human, a human does not know the exact distance like LIDARs. This thesis aims to create an AI that can drive like a human. Humans see an image when driving. Using only a camera was chosen to do this thesis, and the main AI algorithm is a double deep Q-network. When using a double deep Q-network, it requires a simulator to train the network. It will be hard if the network is trained in the real environment due to giving a reward. Training the network in a simulator will not be able to apply to a real car because the network has never seen a real environment before. A semantic segmentation was chosen to solve the problem. There are several semantic segmentation networks. PSPNet is the best choice to use in this case because it can provide quality segmented images, and it can run in real-time. In the end, a golf cart was used to drive in the left lane of a road. The golf cart has been controlled successfully. |
| Year | 2021 |
| Type | Thesis |
| School | School of Engineering and Technology (SET) |
| Department | Department of Industrial Systems Engineering (DISE) |
| Academic Program/FoS | Industrial Systems Engineering (ISE) |
| Chairperson(s) | Mongkol Ekpanyapong |
| Examination Committee(s) | Manukid Parnichkun;Dailey, Matthew N. |
| Scholarship Donor(s) | His Majesty the King's Scholarships (Thailand) |
| Degree | Thesis (M. Eng.) - Asian Institute of Technology, 2021 |