Cyclist behavior: a survey on characteristics and trajectory modeling

Providing realistic simulators is critical to the development of automated vehicles (AVs). A main aspect of simulation is the validation and verification of the driving function in scenarios with vulnerable road users (VRUs), such as cyclists. Modeling VRUs is challenging because their behavior is highly dynamic and the number of possible trajectories is much higher than for vehicles. Their behavior depends on many factors, such as group dynamics, the actions of other road users, and the environment. All of this can lead to unexpected and often irregular behavior. In particular, behavior that does not conform to the rules is difficult to model. This review first provides an overview of the common (mis)behavior of cyclists, statistics of their occurrence, and observable dependencies between misbehavior, personal data, and infrastructure, found in various studies. We also provide an overview of existing methods for modeling cyclist behavior and outline their limitations. Trajectory modeling is mostly dominated by traditional rule-based models such as cellular automata models. In general, existing approaches focus on modeling the standard trajectory and do not cover misbehavior at all, although misbehavior is frequently exhibited by cyclists. We have also found that other aspects of behavior, such as hand gestures, are rarely considered to date.