Lane Change Decision Making for Autonomous Vehicles by using Adversarial Learning Methods
Aytuğ Onurhan Efil* and Mustafa Doğan
ABSTRACT
The primary purpose of this thesis is to analyze the effectiveness of generative adversarial imitation learning and adversarial inverse reinforcement learning in typical highway driving scenarios. The current consensus within the literature indicates that data-based techniques such as IL and IRL when combined with adversarial approaches can produce satisfactory results for autonomous driving on highways. Hence, the central hypothesis is: Can data driven reinforcement learning through adversarial inverse reinforcement learning or generative adversarial imitation learning produce reasonable behavior in a highway driving environment? This research adds to the knowledge base by combining inverse reinforcement learning and imitation learning using generative adversarial network for autonomous driving in highway driving.
