Impact of jaywalking on pedestrian interaction behavior: A multiagent Markov Game-based analysis

Area of Science:

• Traffic Safety
• Behavioral Modeling
• Artificial Intelligence

Background:

• Jaywalking is a significant safety concern in busy traffic environments.
• Existing pedestrian models often fail to account for jaywalking behavior and its impact on crash risk.
• Drivers face unexpected interactions and must take evasive actions due to jaywalkers.

Purpose of the Study:

• To model road user behavior in jaywalking scenarios at signalized intersections using Multiagent Adversarial Inverse Reinforcement Learning (MAAIRL).
• To represent the dynamic decision-making strategies of pedestrians and drivers in jaywalking situations.
• To develop reward functions and optimal policies for improved microsimulation models.

Main Methods:

• Utilized a Markov game framework with MAAIRL to analyze pedestrian and driver interactions.
• Obtained reward functions to infer behaviors and optimal policies for decision-making.
• Compared behavioral patterns and safety metrics between jaywalking and non-jaywalking scenarios.

Main Results:

• Jaywalking pedestrians displayed erratic movements, higher acceleration, and unpredictable paths.
• Non-jaywalking pedestrians exhibited more predictable behavior with larger distances from vehicles.
• Jaywalking scenarios resulted in reduced time-to-collision (TTC) and post-encroachment time (PET), indicating higher crash risks.

Conclusions:

• The MAAIRL model successfully learned distinct behaviors of both jaywalking and non-jaywalking pedestrians.
• Advanced pedestrian simulation models must incorporate jaywalking behavioral patterns for comprehensive safety analysis.
• This framework has the potential to model complex real-world traffic scenarios and enhance safety.