A Study on the Effectiveness of A2C and A3C Reinforcement Learning in Parking Space Search in Urban Areas Problem

Reinforcement learning (RL) helps to select a strategy to execute by gradually predicting and learning according to the reward or punishment feedback given by the environment after selecting a particular strategy to optimize the benefits. The advantage of this model-free method is that it does not need to understand the environment, nor does it take a long time to build a model, but based on what the environment gives, wait for feedback, and take the next step based on the feedback. Reinforcement learning is also suitable for immediate problem-solving applications. This research uses reinforcement learning to solve the problem of searching for parking spaces in urban areas quickly. The proposed method only needs to set up sensors at the road intersections to sense the vehicles and count the number of vehicles passing through, and the probability of parking vacancy can be estimated based on the length of the road and the number of vehicles entering and exiting the road in a specific time interval. Then through the evaluation results of the policy-based A2C (Advantage Actor-Critic) and A3C (Asynchronous Advantage Actor-Critic), it provides vehicles with the most likely parking routes suggestions. This research uses the traffic flow and parking information of each time period in the road segment of the Taipei city. At last, we compare the expected searching time of A2C and A3C reinforcement learning in the parking space search problem in urban areas.