FOSS4G-Asia 2023 Seoul

To improve the safety of autonomous driving, ample data is required. However, obtaining driving data takes considerable manpower, and driving an autonomous vehicle is dangerous. Recently, vehicle driving simulations have been efficiently researched in the field of autonomous driving as a solution to this problem. However, these simulations are difficult to obtain for public users, and users must pay expensive license fees. Moreover, existing vehicle driving simulations can only be used in a limited environment because users have limited modification rights. Therefore, we want to develop a simulator with better accessibility than paid simulators and establish a simulation each user can acquire data in the customized environment. It will include the core functions of commercial simulators, such as setting the vehicle's route, creating a realistic environment, and acquiring the vehicle's sensor data. This simulator, implemented as a "toolchain," incorporates open-source platforms.
In this study, we will integrate the open-source platforms 'Vehicle Route Planning Simulation Platform', '3D Vehicle Driving Simulation Platform', and '3D Road Modeling Tool'. The 'Vehicle Route Planning Simulation Platform' is specialized in planning the route of a vehicle, the '3D Vehicle Driving Simulation Platform' is developed to support autonomous driving systems so it is easy to obtain vehicle sensor data, and the '3D Road Modeling Tool' helps to build a realistic environment. Therefore, we adopted these platforms. Vehicle driving simulation have essential elements.
We need a Vector Map the vehicle references to drive and a 3D model to visualize it. Based on the vector map, the scenario of the vehicle is determined and the environment similar to the real world is created with the 3D model. To implement the actual vector map, we first download the road network in OSM format from Openstreetmap. And the road network is converted to a road network used by the 'Vehicle Route Planning Simulation Platform'(network). The network generates vehicle scenario(route) and 3D road data with road shape and properties through the 'Vehicle Route Planning Simulation Platform'. The 3D road data has a real-world environment by creating buildings and structures, vegetation, and the slope of the road using the 3D Road Modeling Tool. Import the previously created network file, route file, and 3D road data into the '3D Vehicle Driving Simulation Platform', and you can see the vehicles driving on a set route in the 3D simulator. Additionally, We define Camera, LiDAR, and GNSS sensors and assign the location and properties of the sensors. And we can obtain customized data. Using the method described above, we implemented a simulator based on the environment near the main gate of Seoul National University and obtain a camera, LiDAR, and GNSS data from 1000 vehicles at a time. The vehicle driving simulator creates a virtual mirror world of the real road. Moreover, simulators with open source can interact with various developers and improve rapidly, which is an opportunity to expand the simulator market. Furthermore, this research is expected to help develop algorithms to improve the location information of vehicles.