Open Collections

Mah'd Alsaleh, Rushdi

University of British Columbia

Understanding and modeling cyclist and pedestrian dynamics and their microscopic interaction behaviour in shared space facilities are crucial for several applications, including safety and performance evaluations. Recently, a few studies have developed models to simulate road user interactions in shared space facilities. However, existing models suffer from several shortcomings and show significant discrepancies with real-world behaviour. As such, this thesis presents a novel microsimulation-oriented framework for modeling cyclist and pedestrian interactions in such facilities. Advanced Artificial-Intelligent techniques were used to model road users' behaviour and their interactions as reward-based intelligent agents. This thesis bridges the gap in modeling road user interactions by accounting for their rationality, intelligence, and sequential decision-making process by implementing the Markov Decision Process (MDP) modeling framework. Furthermore, this thesis proposes a multi-agent modeling framework to model cyclist and pedestrian interactions in shared spaces. Unlike the traditional game-theoretic framework that models multi-agent systems as a single time-step payoff, the proposed approach is based on Markov Games (MG), which models road users' sequential decisions concurrently. Moreover, this thesis investigates the ability of different equilibrium behavioral theories (i.e., Nash-Equilibrium (NE) and Logistic-Stochastic-Best-Response-Equilibrium (LSBRE)) in predicting road user operational-level decisions and evasive-action mechanisms. Road user trajectories from three shared space facilities located in Vancouver, Canada, and New York City, USA, were extracted by means of computer vision algorithms. Single and multi-agent inverse reinforcement learning approaches were utilized to estimate road user reward functions using examples of their demonstration (i.e., trajectories). Reward function weights infer road users' goals and preferences and can form the key component in developing agent-based microsimulation models. Single-agent and multi-agent simulation platforms were developed, relying on deep reinforcement learning approaches, to emulate and validate road user interactions in shared spaces. The utilized multi-agent modeling approaches led to a significantly more accurate prediction of road user behaviour and their evasive action mechanisms. Moreover, the recovered reward functions based on the single-agent modeling approach failed to capture the equilibrium solution concept compared to the multi-agent approach. This thesis determines a behavior-based consistent paradigm to model equilibrium in multi-agent transportation systems, such as road user interactions in shared space facilities.

Text

2021-12-02

Attribution-NonCommercial-NoDerivatives 4.0 International

http://hdl.handle.net/2429/80313

Civil Engineering

University of British Columbia

UBCV

http://creativecommons.org/licenses/by-nc-nd/4.0/