Evorobotpy3: a flexible and easy-to-use simulation tool for Evolutionary Robotics

The paper presents "evorobotpy3", a flexible and easy-to-use Python-based simulation environment for implementing and testing various algorithms, addressing diverse benchmark problems, and supporting different neural network controllers. Although existing simulators can model real robots accurately, they are often limited in generalizability and tailored to specific domains and applications. In contrast, evorobotpy3 is designed to be extensible and adaptable, not only by allowing customization for different domains but also by inherently incorporating the operational principles of evolutionary algorithms by design. To demonstrate these capabilities, we evaluated the OpenAI-ES algorithm on a series of benchmark tasks, including Pybullet locomotion tasks and classic control problems such as pole balancing. This study underscores the potential of evorobotpy3 as a powerful and extensible tool for robotics and artificial intelligence research.