One of the most challenging problems with developing technology for autonomous vehicles is access to annotated data for supervised learning and simulation environments for reinforcement learning. In this project, we want to develop a simulation tool based on machine learning, tailored specifically for the domain of autonomous vehicles. The tools, known as SweSim4AD and RealSim4AD, will learn to translate synthetic data from a simulation to more realistic data using unsupervised machine learning. The tool for this project will reflect Swedish urban areas, country road and highways and capable of both be used to learn intelligent behaviour by agents through reinforcement learning, as well as generate training data in the form of depth maps, bounding boxes, object segmentation and other common targets for supervised machine learning. We hope this project will take some great steps towards creating the world´s best environment for developing autonomous vehicle technology.
vid SAFER, The Vehicle and Traffic Safety Centre
Funding Chalmers participation during 2017–2019