Most conventional pedestrian models have been designed for high-density, chaotic situations such as indoor emergency evacuation scenarios. For this reason, they tend to exhibit relatively short-sighted collision avoidance. Some newmodels, however, propose the use of parameters such as time to interaction and the time derivative of the bearing angle to the interacting pedestrian to detect and avoid potential collisions further in advance. The use of these parameters is intuitive, and themodels using them provide impressively realistic results, however the core hypotheses that underpin these new models has yet to be thoroughly tested on real-world data. In this paper, therefore, we evaluate the power of these and other parameters for the prediction of pedestrian trajectories in the context of street traffic using the InD naturalistic driving dataset. We find that, while these parameters do present a useful way to visualize pedestrian interactions which can reveal many interesting behavioral patterns, the specific control mechanisms underlying the aforementioned models do not appear tomatch the real-world data to a statistically significant degree. In addition, we find evidence that both the side from which another pedestrian is approaching and their orientation relative to the ego pedestrian also play a significant role in determining pedestrian behavior. In particular, we observe a slight preference to “keep right” in head-on interactions, as well as a slightly higher tendency to “yield” to pedestrians coming from the right side at a right angle than to those coming from the left side.     «

Most conventional pedestrian models have been designed for high-density, chaotic situations such as indoor emergency evacuation scenarios. For this reason, they tend to exhibit relatively short-sighted collision avoidance. Some newmodels, however, propose the use of parameters such as time to interaction and the time derivative of the bearing angle to the interacting pedestrian to detect and avoid potential collisions further in advance. The use of these parameters is intuitive, and themodels us...     »