Smaller medial-lateral separation between approaching pedestrians and lack of an additional interferer predicted higher likelihood of path deviation, while presence of an additional interferer, smaller group size, and pedestrian distractedness were associated with greater medial-lateral separation at time of crossing.
Key Findings
Results
Smaller medial-lateral separation between approaching parties predicted a higher likelihood of path deviation for one or both pedestrians.
Analysis was conducted on unscripted pedestrian interactions in a real-world busy urban path environment.
Deep learning algorithms were applied to detect and extract pedestrian walking trajectories.
Multiple regression analysis was used to determine which factors influenced collision avoidance behaviours.
Medial-lateral separation was measured between a single pedestrian and an approaching group.
Results
Lack of an additional interferer (a pedestrian not directly involved in the interaction) in the area predicted a higher likelihood of path deviation for one or both parties.
The presence or absence of an additional interferer was characterized by unbiased raters reviewing video footage.
The additional interferer was defined as a pedestrian not involved in the primary interaction.
This factor was identified through multiple regression analysis as a significant predictor of path deviation likelihood.
The finding suggests that third-party pedestrians in the vicinity influence collision avoidance decisions.
Results
The presence of an additional interferer in the area was associated with greater medial-lateral separation at the time of crossing.
Medial-lateral separation at time of crossing was used as a dependent variable in regression analyses.
The presence of an additional interferer was one of three factors found to significantly predict greater separation at crossing.
This finding indicates that third-party pedestrians affect not only the likelihood but also the extent of path deviations.
The direction of this effect (greater separation with an interferer present) contrasts with its effect on likelihood of deviation.
Results
Smaller group size was associated with greater medial-lateral separation at the time of crossing.
Group size was one of the hypothesized factors tested in the multiple regression analysis.
The study involved interactions between a single pedestrian and a group of pedestrians on a busy urban path.
Smaller groups were associated with pedestrians achieving greater lateral separation when paths crossed.
Group size did not emerge as a significant predictor of the likelihood of path deviation, only of the extent.
Results
Pedestrian distractedness was associated with greater medial-lateral separation at the time of crossing.
Distractedness was characterized by unbiased raters reviewing video footage of pedestrian interactions.
Distractedness was one of the factors hypothesized to influence collision avoidance behaviours.
It was associated with greater (not lesser) lateral separation at the moment paths crossed.
Distractedness did not significantly predict the likelihood of path deviation in the regression model.
Results
Pedestrian age and constraints to mobility (holding or pushing an object) were hypothesized factors that did not emerge as significant predictors in the regression analyses.
Both pedestrian age and mobility constraints were included as candidate predictors in the multiple regression model.
Neither factor was reported among the significant predictors of path deviation likelihood or medial-lateral separation at crossing.
The study was conducted in a natural environment with unscripted pedestrian behaviours.
Unbiased raters characterized pedestrian interactions including mobility constraints and age estimation.
Methods
The study used deep learning algorithms applied to video footage of unscripted pedestrian interactions on a busy urban path to extract walking trajectories.
Videos captured natural, real-world pedestrian walking behaviours without scripting or participant awareness.
Deep learning algorithms were used to detect pedestrians and extract their walking trajectories from the video data.
Unbiased raters independently characterized pedestrian interactions for variables such as distractedness, mobility constraints, and presence of additional interferers.
Multiple regression analysis was the primary statistical approach used to identify contributing factors to collision avoidance behaviours.
What This Means
This research examined how people walking in groups and individual pedestrians avoid colliding with each other in a real, busy outdoor environment. Instead of using a laboratory or scripted scenarios, researchers recorded natural pedestrian interactions on an urban walking path and used artificial intelligence to track people's movements. They then analyzed what factors made it more or less likely that someone would swerve or step aside to avoid a collision, and how much space people ended up with when their paths crossed.
The study found that when people were closer together as they approached each other, they were more likely to change their path to avoid a collision. Interestingly, having a third, uninvolved pedestrian nearby also influenced behavior — its absence made people more likely to deviate, while its presence was linked to people having more space between them when they passed. Smaller groups, and whether someone was distracted (for example, looking at a phone), were also linked to greater side-to-side spacing when paths crossed. Factors like age and whether someone was carrying or pushing something did not significantly predict these behaviors.
This research suggests that collision avoidance between pedestrians is shaped by multiple interacting factors beyond just the direct path of the people involved — nearby bystanders, group size, and attention all play roles. These findings could be useful for improving computer models that simulate crowd movement and for programming robots or autonomous vehicles to navigate more safely and naturally around groups of people in public spaces.
Check Your Own Numbers
Upload your bloodwork. We'll cross-reference your results against this study and 4,700 others.
Nikmanesh M, Cinelli M, Marigold D. (2026). Factors that contribute to collision avoidance behaviours involving a single pedestrian versus a group of pedestrians in a natural environment.. Human movement science. https://doi.org/10.1016/j.humov.2026.103485