Interface OpenSCENARIO for CARLA simulator using ROS control

Abstract

The potential impact self-driving cars could have in our society's future inspired many researchers to switch their focus to further develop this technology. In the recent years its popularity and financing has increased substantially with startups and some major car companies getting involved. Regardless of this trend, the technology still faces many challenges ahead. The leading advances are based on machine learning techniques, which require training on a variety of situations in real-life or with large datasets. However, security liabilities and ethical questions prevent these vehicles from performing freely on the public roads, halting the development progress. Urban driving simulators present themselves as the solution, allowing to train ADSs in realistic and complex urban scenarios while reducing costs and without endangering human lives. Many urban driving simulators are available nowadays but all of them developed their own interface and APIs resulting in a largely non-standard industry. This thesis proposes the development of a communication layer for CARLA simulator providing a standard method to interface with the control algorithms. This implementation, although tuned specifically for this simulator, will allow full interaction with the simulator using only ROS and OpenScenario standard.