Bias in Automated Driving Systems (BADS)

Abstract

Automated Driving Systems do about 100% more kilometers in simulation than in physical space. This raises questions about the quality of data sets and data records shaping the abilities of these driving systems. Currently there are about 40 accessible public data sets, which have already been briefly analyzed by the project researchers. This project aims at a systematic evaluation of this material at this early stage of technological development.

Outcome Summary

The research conducted by the Bias in Automated Driving Systems (ADS) team led to a comprehensive review of 38 international datasets on ADS scenarios and sequences. This led to the development of a WWTF application for the Digital Humanism 2020 call. The aim was to explore the future of ADS learning environments in the context of Greater Vienna. The interdisciplinary research team aimed to answer questions about learning environments and the systematic biases in them, investigate the potential future use of the system in the context of different user groups, and develop guidelines to compensate for the biases in the training data.

To address these complex research questions, a mixed-methods approach was proposed, consisting of a comprehensive analysis of existing datasets related to the use of ADS in both simulated and real environments, expert interviews with those working directly with ADS systems, and participatory methods in collaboration with the aspern.mobil LAB.