Benchmarking Autonomous Driving Systems using a Modular and Scalable Simulation Environment

Autonomous Driving technology has advanced significantly over the past decades driven by key developments in Machine Learning, Computer Vision and Control Systems. Despite its potential to reduce traffic accidents, statistically validating the safety of autonomous vehicles remains a significant challenge that can only be addressed through simulation. Simulation environments are crucial for benchmarking and testing Autonomous Driving systems in a controlled and repeatable manner, but researchers often use different simulation strategies for each study, making it difficult to compare results and draw consistent conclusions. Inspired by the standardization efforts seen in platforms like OpenAI Gym, this paper proposes a modular simulation environment that accurately models path planning, compares control strategies, and simulates sensor noise. It aims to enhance the consistency and comparability of research by providing a platform to benchmark and test Autonomous Driving systems in a flexible, and accessible manner.