IoT Based Smart Parking Systems in Unity With ML-agents Toolkit

Modern cities face problems with limited parking space, which requires effective optimization to reduce congestion. This paper presents an up-to-date approach to solving the problem of parking in restricted space conditions using the Unity Machine Learning Agents Toolkit (ML-Agents). This toolkit is based on artificial neural networks. It opens up wide opportunities for training agents capable of performing tasks in real time to optimize the use of parking areas and reduce the time to search for free spaces in real-time. Data transmission from sensors in Internet of Things (IoT) systems plays a key role in optimizing the use of parking spaces and increasing driver comfort. For this purpose, modern IoT technologies are integrated into the system, which allows the effective solving of parking problems in limited urban space conditions. The concept of an intelligent parking system is presented, which uses sensor technologies and machine learning algorithms to increase the efficiency of the process and driver comfort. Despite the high costs of installation and maintenance of traditional systems, the use of the Unity game engine and ML-Agents allows you to create training environments for preliminary testing, debugging, and optimization of algorithms. In the course of the research, a machine learning model was developed and analyzed using the Proximal Policy Optimization algorithm, which allowed us to reproduce various agent training scenarios. This contributed to the acceleration and stabilization of the training process, as well as the establishment of optimal model parameters based on the analysis of key performance metrics. The results of testing and comparative analysis confirm the prospects of the proposed approach in the field of autonomous car parking.