CarAware: A Deep Reinforcement Learning Platform for Multiple Autonomous Vehicles Based on CARLA Simulation Framework

• Published in: 2023 8th International Conference on Models and Technologies for Intelligent Transportation Systems (MT-ITS) pp. 1 - 6
• Main Authors: Araujo, Tulio Oliveira, Netto, Marcio Lobo, Justo, Joao Francisco
• Format: Conference Proceeding
• Language: English
• Published: IEEE 14-06-2023
• Subjects: Connected Autonomous Vehicles; Data integration; Data models; Deep learning; Deep Reinforcement Learning; Focusing; Location awareness; Reinforcement learning; Simulation Framework; Training
• DOI: 10.1109/MT-ITS56129.2023.10241376

To facilitate studies in Deep Reinforcement Learning (DRL) and autonomous vehicles, we present the CarAware framework 1 for detailed multi-agent vehicle simulations, which works together with the open-source traffic simulator CARLA. This framework aims to fill the gap identified in currently available CARLA DRL frameworks, often focused on the perception and control of a single vehicle. The new framework provides baselines for training DRL agents in scenarios with multiple connected autonomous vehicles (CAVs), focusing on their sensors' data fusion for objects' localization and identification. These features and tools allow studying many different DRL strategies and algorithms, applied for multi-vehicle sensors' data fusion and interpretation. 1 https://github.com/tulioaraujoMG/CarAware