Logo Logo
Switch Language to German
Hollaender, Kai; Ou, Changkun; Schellenberg, Luca; Butz, Andreas (2020): All Fun and Games: Obtaining Critical Pedestrian Behavior Data from an Online Simulation. In: Chi'20: Extended Abstracts of the 2020 Chi Conference on Human Factors in Computing Systems, LBW091
Full text not available from 'Open Access LMU'.


Automated cars will need to observe pedestrians and react adequately to their behavior when driving in urban areas. Judging pedestrian behavior, however, is hard. When approaching it by machine learning methods, large amounts of training data is needed, which is costly and difficult to obtain, especially for critical situations. In order to provide such data, we have developed an online game inspired by Frogger, in which players have to cross streets. Accidents and critical situations are a natural part of the data produced in such a way without anybody getting hurt in reality. We present the design of our game and an analysis of the resulting data and its match to real world behavior observed in previous work. We found that behavior patterns in real and virtual environments correlated and argue that game data could be used to train machine learning algorithms for predicting real pedestrians' walking trajectories when crossing a road. This approach could be used in future automated vehicles to increase pedestrian safety.