In this thesis I present a fast, efficient, and production friendly method to generate a crowd of fully articulated characters. A wide variety of characters can be created from a relatively few base models. The models that are generated are anatomically different from each another, while maintaining the same topology. They all have individual characteristics and features, that distinguish them from the others in the crowd. This method is easily adaptable to different kinds of characters, from hyper-realistic characters to highly stylized characters, and from human characters to insects like spiders. The crowd character models generated by this method are fully articulated and are ready to be animated.

京都大学 / 新制・課程博士 / 博士(情報学) / 甲第24726号 / 情博第814号 / 新制||情||137(附属図書館) / 京都大学大学院情報学研究科知能情報学専攻 / (主査)教授 西野 恒, 教授 河原 達也, 教授 神田 崇行, 准教授 延原 章平 / 学位規則第4条第1項該当 / Doctor of Informatics / Kyoto University / DFAM

We study the mean-field limit of a particle-based system modeling the behavior of many indistinguishable pedestrians as their number increases. The base model is a modified version of Helbing's social force model. In the mean-field limit, the time-dependent density of two-dimensional pedestrians satisfies a four-dimensional integro-differential Fokker-Planck equation. To approximate the solution of the Fokker-Planck equation we use a time-splitting approach and solve the diffusion part using a Crank-Nicholson method. The advection part is solved using a Lax-Wendroff-Leveque method or an upwind Backward Euler method depending on the advection speed. Moreover, we use multilevel Monte Carlo to estimate observables from the particle-based system. We discuss these numerical methods, and present numerical results showing the convergence of observables that were calculated using the particle-based model as the number of pedestrians increases to those calculated using the probability density function satisfying the Fokker-Planck equation.

A multi-scale agent-based simulation framework is firstly proposed to analyze pedestrian delays at signalized crosswalks in large urban areas under different conditions. The aggregated-level model runs under normal conditions, where each crosswalk is represented as an agent. Pedestrian counts collected near crosswalks are utilized to derive the binary choice probability from a utility maximization model. The derived probability function is utilized based on the extended Adam's model to estimate an average pedestrian delay with corresponding traffic flow rate and traffic light control at each crosswalk. When abnormality is detected, the detailed-level model with each pedestrian as an agent is running in the affected subareas. Pedestrian decision-making under abnormal conditions, physical movement, and crowd congestion are considered in the detailed-level model. The detailed-level model contains two sub-level models: the tactical sub-level model for pedestrian route choice and the operational sub-level model for pedestrian physical interactions. The tactical sub-level model is based on Extended Decision Field Theory (EDFT) to represent the psychological preferences of pedestrians with respect to different route choice options during their deliberation process after evaluating current surroundings. At the operational sub-level model, physical interactions among pedestrians and consequent congestions are represented using a Cellular Automata model, in which pedestrians are allowed biased random-walking without back step towards their destination that has been given by the tactical sub-level model. In addition, Dynamic-Data-Driven Application Systems (DDDAS) architecture has been integrated with the proposed multi-scale simulation framework for an abnormality detection and appropriate fidelity selection (between the aggregate level and the detailed level models) during the simulation execution process. Various experiments have been conducted under varying conditions with the scenario of a Chicago Loop area to demonstrate the advantage of the proposed framework, balancing between computational efficiency and model accuracy. In addition to the signalized intersections, pedestrian crossing behavior under unsignalized conditions which has been recognized as a main reason for pedestrian-vehicle crashes has also been analyzed in this dissertation. To this end, an agent-based model is proposed to mimic pedestrian crossing behavior together with drivers' yielding behavior in the midblock crossing scenario. In particular, pedestrian-vehicle interaction is first modeled as a Two-player Pareto game which develops evaluation of strategies from two aspects, delay and risk, for each agent (i.e. pedestrian and driver). The evaluations are then used by Extended Decision Field Theory to mimic decision making of each agent based on his/her aggressiveness and physical capabilities. A base car-following algorithm from NGSIM is employed to represent vehicles' physical movement and execution of drivers' decisions. A midblock segment of a typical arterial in the Tucson area is adopted to illustrate the proposed model, and the model for the considered scenario has been implemented in AnyLogic® simulation software. Using the constructed simulation, experiments have been conducted to analyze different behaviors of pedestrians and drivers and the mutual impact upon each other, i.e. average pedestrian delay resulted from different crossing behaviors (aggressive vs. conservative), and average braking distance which is affected by driving aggressiveness and drivers' awareness of pedestrians. The results look interesting and are believed to be useful for improvement of pedestrians' safety during their midblock crossing. To the best of our knowledge, the proposed multi-scale modeling framework for pedestrians and drivers is one of the first efforts to estimate pedestrian delays in an urban area with adaptive resolution based on demand and accuracy requirement, as well as to address pedestrian-vehicle interactions under unsignalized conditions.

Autonomous robots interacting with humans in a social setting must consider the social-cultural environment when pursuing their objectives. Thus the social robot must perceive and understand the social cultural environment in order to be able to explain and predict the actions of its human interaction partners. This dissertation contributes to the emerging field of human-robot interaction for social robots in the following ways: 1. We used the social calculus technique based on culture sanctioned social metrics (CSSMs) to quantify, analyze and predict the behavior of the robot, human soldiers and the public perception in the Market Patrol peacekeeping scenario. 2. We validated the results of the Market Patrol scenario by comparing the predicted values with the judgment of a large group of human observers cognizant of the modeled culture. 3. We modeled the movement of a socially aware mobile robot in a dense crowds, using the concept of a micro-conflict to represent the challenge of giving or not giving way to pedestrians. 4. We developed an approach for the robot behavior in micro-conflicts based on the psychological observation that human opponents will use a consistent strategy. For this, the mobile robot classifies the opponent strategy reflected by the personality and social status of the person and chooses an appropriate counter-strategy that takes into account the urgency of the robots' mission. 5. We developed an alternative approach for the resolution of micro-conflicts based on the imitation of the behavior of the human agent. This approach aims to make the behavior of an autonomous robot closely resemble that of a remotely operated one.