Perpustakaan Digital ITB

An autonomous vehicle (AV) is a vehicle that can operate fully autonomously without any human intervention. AV popularity has been rising due to the advancements in sensors, computing power, and artificial intelligence. Autonomous parking is part of this autonomous driving technology, which is becoming more important because every vehicle needs to park. One of the control methods for autonomous driving is deep reinforcement learning. This research aims to improve the autonomous control model from the previous study and compare the overall performance of the two. The research started by creating the simulation environment in CARLA. The previous study used the deep deterministic policy gradient algorithm (DDPG). Meanwhile, this study implemented a better algorithm called twin-delayed deep deterministic (TD3). Following this, the research continued by creating two models, one that uses DDPG and the other that uses TD3. Afterwards, both models are trained for backward parallel parking and perpendicular parking. Then, they are tested to see the training results. The model training and the testing results are then evaluated to see which model is better. The result shows that for the parallel parking scenario, the model that uses the TD3 algorithm performs better. In contrast, for the backward perpendicular parking scenario, both of the models performed identically while being tested.