30th Signal Processing and Communications Applications Conference, SIU 2022, Safranbolu, Türkiye, 15 - 18 Mayıs 2022
© 2022 IEEE.Deep reinforcement learning has achieved human-level and even beyond performance on complex tasks like Atari games and Go. However, this performance is not easy to adapt to autonomous driving since real world state spaces are extremely complex and have continuous action spaces. Besides, autonomous driving tasks often require decision making under uncertainty. Hence, the autonomous driving problem can be formulated as a partially observable Markov decision process (POMDP).In this paper, we propose a new approach to solve the autonomous driving problem based on decision making under uncertainty as a partially observable Markov decision process, using Guided Soft Actor-Critic (Guided SAC). Self driving car has been trained for the scenario where it encountered with a pedestrian crossing the road. Experiments show that the control agent exhibits desirable control behavior and performed close to the fully observable state under various uncertainty situations.