CROWD-RL: Ambiente Multiagente para Simulação de Fluxo de Atendimento

Enterprises and public institutions face challenges in planning their physical spaces, organizing queues, and managing attendants to ensure efficient customer service. In this context, this article presents the development of a Python library that enables the modeling of real physical environments, along with the application of multi-agent reinforcement learning algorithms, to generate simulations in which various agents navigate the modeled location and go through queues and attendants. To achieve this objective, a literature review was conducted, reinforcement learning techniques were explored, and the characteristics to be included in the environment and agents were defined. All this culminated in the selection of the Python programming language with the PettingZoo API and the development of the configurable library "Crowd- rl" with the capacity to model locations, entrances, exits, attendants, queues, and agents. Simulations of real environments, such as bus station counters, supermarkets, and government service centers, were carried out.