Presentation
Logical Reasoning Enhances Detection of Pedestrian Behavior Changes
DescriptionImitating driver’s capability in pedestrian prediction is essential for improving AI-enabled autonomous driving. Previous research suggests that people have the status quo tendency to predict continued pedestrian actions, reflecting a reliance on intuitive reasoning. However, learning only bottom-up decisions may reduce decision accuracy during pedestrian interactions.
To better imitate top-down decisions, this study examines how logical reasoning influences people’s prediction of pedestrian behavior changes.
A total of 163 participants recruited through crowdsourcing platforms completed two experiments using 70 pedestrian videos, including Transition Cases (with behavior changes) and Continued Cases (without changes). In Experiment-Continued (Exp-C), participants were prompted to use intuitive reasoning by judging whether the pedestrian would continue current action. In Experiment-Transition (Exp-T), they were encouraged to use logical reasoning by judging whether the pedestrian would switch actions. We hypothesized that relative to Exp-C, logical reasoning (Exp-T) would enhance participants’ ability to detect behavior changes using behavior and contextual cues—resulting in a higher prediction rate of changes on Transition Cases than Continued Cases.
Results show that the increase in transition prediction rate from Exp-C to Exp-T was significantly greater in Transition Cases than in Continued Cases (p = 0.015), suggesting that logical reasoning was effective when more cues were present to predict behavior transitions.
These findings indicate that encouraging logical reasoning helps participants recognize potential changes in pedestrian crossing intentions by reducing intuitive bias and enhancing judgment in dynamic environments. By modeling this reasoning process, AI systems can better reflect human social cognition and improve safety in pedestrian-vehicle interactions.
To better imitate top-down decisions, this study examines how logical reasoning influences people’s prediction of pedestrian behavior changes.
A total of 163 participants recruited through crowdsourcing platforms completed two experiments using 70 pedestrian videos, including Transition Cases (with behavior changes) and Continued Cases (without changes). In Experiment-Continued (Exp-C), participants were prompted to use intuitive reasoning by judging whether the pedestrian would continue current action. In Experiment-Transition (Exp-T), they were encouraged to use logical reasoning by judging whether the pedestrian would switch actions. We hypothesized that relative to Exp-C, logical reasoning (Exp-T) would enhance participants’ ability to detect behavior changes using behavior and contextual cues—resulting in a higher prediction rate of changes on Transition Cases than Continued Cases.
Results show that the increase in transition prediction rate from Exp-C to Exp-T was significantly greater in Transition Cases than in Continued Cases (p = 0.015), suggesting that logical reasoning was effective when more cues were present to predict behavior transitions.
These findings indicate that encouraging logical reasoning helps participants recognize potential changes in pedestrian crossing intentions by reducing intuitive bias and enhancing judgment in dynamic environments. By modeling this reasoning process, AI systems can better reflect human social cognition and improve safety in pedestrian-vehicle interactions.
Event Type
Late Breaking Results
TimeFriday, October 17th8:20am - 8:40am CDT
LocationGrand C/D South
General Sessions
Similar Presentations