Simulating multi-agent narratives for pre-occupancy evaluation of architectural designs

Davide Schaumanna,, Nirit Putievsky Pilosofb, Hadas Sopherb, Jacob Yahavb, Yehuda E. Kalayb

Department of Computer Science, Rutgers University, New Brunswick, NJ, United States of America Faculty of Architecture & Town Planning, Technion – Israel Institute of Technology, 3200003, Haifa, Israel

A R T I C L E I N F O A B S T R A C T

Keywords:

Multi-agent narratives

Pre-occupancy evaluation

Human behavior simulation

Architectural design options

Multi-agent systems

 

Simulating and evaluating the impact that a building design might produce on its prospective occupants is a key challenge in architectural design. Prior work demonstrated the capabilities of narrative-based modeling to coordinate the collaborative behavior of virtual occupants. In this work, we aim to demonstrate the scalability and applicability of narrative-based modeling to support the pre-occupancy evaluation of alternative design options in complex real-world hospital facilities. To do so, we developed a narrative-based pre-occupancy evaluation platform that extends pre-existing narrative-based capabilities with (a) a newly developed library space, actor, activities, and narrative entities that support the simulation of real-world human behavior patterns while accounting for the impact that a building design produces on how the patterns unfold, and (b) a newly integrated evaluation module able to generate and visualize numerical data-logs and spatiotemporal data-maps of key performance indicators in hospital settings. We applied the platform to conduct a comparative pre-occupancy evaluation of two different architectural designs for an outpatient ophthalmology clinic. Results demonstrate the scalability and applicability of narrative-based modeling to help design stakeholders visualize and analyze how design decisions may impact future building operations in outpatient clinics.

1. Introduction

One of the major challenges in architectural design is to predict how a building will perform with respect to its future human inhabitants. In recent years, buildings have become increasingly more complex in terms of size, cost of operations, spatial organization, function, and number and types of users. Analyzing building-user interactions at the time a building is being designed is thus becoming increasingly more challenging even for the most skillful architects. As result, much uncertainty is left in the overall performance of the designed architectural products, which may contribute to a dissonance between the expected and the actual users' behavior in built environments [1,2]. The stakes for designing a successful environment, however, are very high. A building that successfully meets its user needs holds promise to maximize users' productivity and satisfaction while guaranteeing operational efficiency as well as an optimized space utilization. Oppositely, a building that does not support its users' needs may lead to lack of functionality, high costs, reduction of users' productivity [2–4].

To predict and evaluate building performance, architects have traditionally used physical and later digital building models. Recent developments in Computer Aided Design (CAD) and Building Information Modeling (BIM) led to the development of a plethora of tools that help architects analyze their designs with respect to a wide range of building performances, mostly in terms of energy [5], structure [6], construction [7] and daylight [8]. Nevertheless, predictions and evaluations of how people will use a building after its construction and occupancy have developed more slowly compared to the aforementioned tools. Computational analyses of this kind require not only a digital model of the building (as the one generated with CAD and BIM tools), but also a model of the building occupants and the dynamic activities they perform in a given space, at a given time. This is indeed a complex endeavor since built environments are inhabited by many users with different roles, goals and needs.

 

To address this issue, multi-agent systems (MAS) have been developed to represent dynamic building-user interactions in built environments. Current MAS approaches are mainly built bottom-up, meaning that each agent is equipped with independent perception-action abilities. These methods focus on linear and straightforward occupancy scenarios such as pedestrian movement under normal or emergency conditions [9–13]. Other approaches, defined as top-down, coordinate agents' behavior by means of fixed activity schedules in office and university buildings [14,15].

Narrative-based modeling has been proposed as a synthesis of bottom-up and top-down approaches to simulate non-linear day-to-day use scenarios in complex settings, like hospitals, where multiple heterogeneous agents are involved in planned and unplanned collaborative activities dynamically unfolding in semantically-rich spaces [16]. Centered on narratives, this approach uses rule-based scripts that direct the collaborative behavior of multiple occupants. Different from other MAS approaches, narratives provide a top-down coordination mechanism to enforce the performing of structured sets of tasks, while allowing for dynamic bottom-up adaptations to dynamically changing social and spatial conditions.

Although prior work demonstrated that narrative-based modeling can simulate planned and unplanned building-human interactions in abstracted hospital settings [16], the scalability and applicability of this model to conduct comparative pre-occupancy evaluations of real-world hospital designs is yet to be demonstrated.

This is an important study since a model's ability to simulate human behavior in real-world settings and applicability to generate and visualize occupancy data tailored to a specific built environment hold promise to help design stakeholders make more informed decisions at the design stage, when issues can be identified and solved before it is too late or costly.

However, this is a challenging task due to the complex nature of hospital settings. These environments are populated by hundreds of occupants with different roles, goals and needs, who are engaged in a variety of task-based collaborative activities that impact (and are impacted by) spatial features as well as the presence and activities of other occupants. In such settings, the mutual relations between building design and occupant behavior can be measured using a variety of key performance indicators (KPIs) representative of a building's operational performance, level of service, occupant satisfaction and overall spatial utilization.

Prior work on narrative-based modeling, however, demonstrated the behavior of a limited number of agents engaged in representative behaviors in an abstracted setting over a short time span. In addition, prior work did not demonstrate the ability of the model to calculate meaningful KPIs that could support pre-occupancy evaluation by different design stakeholders.

In an effort to demonstrate the scalability and applicability of narrative-based modeling for pre-occupancy evaluation of real-world hospital designs, we developed a platform that extends pre-existing narrative-modeling capabilities with two key features aimed at addressing its current shortcomings, namely (a) a novel library space, actor, activities, and narrative entities that support the simulation of realworld human behavior patterns and the impact that a building design produces on how the patterns unfold, and (b) a newly integrated evaluation module able to generate numerical data-logs and spatiotemporal data-maps of key performance indicators tailored to hospital settings. We tested the proposed platform by conducting a comparative pre-occupancy evaluation of two real-world design options for an outpatient ophthalmology clinic.

In the following sections, we first review existing studies on human behavior simulation and their applications for architectural design. Then, we outline our narrative-based pre-occupancy evaluation platform. Following, we present a comparative study between two design options of an ophthalmology outpatient clinic. Finally, we discuss the proposed approach and we outline the advantages, limitations and future directions for this work.

2. Human behavior simulation in architectural design: a review

2.1. Pre-occupancy evaluation in architectural design

During the design process, architects evaluate a wide set of solutions to identify the one(s) that better meet defined goals, while abiding by specific constraints [17,18]. However, because of the wicked nature of the design process, the implications of architectural design decisions are not self-evident in the design phases [19] – they become evident only after a building is built and occupied.

Computational tools have thus been developed to help architects evaluate the implications of architectural design solutions mostly in terms of cost, energy, structure, and daylight. Other approaches investigated the impact of built environments on their human inhabitants. Different from the aforementioned approaches, analyses of this kind are predicated on the understanding of social and behavioral aspects of building occupants, which are hard to measure and quantify since the proposed buildings do not yet exist. This challenge is further exacerbated by the complex nature of the built environments we inhabit, which host many functions and user types with different needs. Hospitals, for instance, are populated by permanent and temporary users that may have conflicting goals and needs: while patients and visitors strive to maximize the interactions with the staff members to improve communication and treatment coordination, such interactions may have detrimental effects on the staff performance since they may produce delays and medical errors [20].

A common practice to analyze the relationship between a built environment and its future building occupants are full-scale mockups. Patient rooms and operating theaters, for example, are built and inhabited to test ergonomic aspects of users' occupancy [21]. While these studies employ real people for evaluation, they can only be applied to small portions of a building. Furthermore, they fail to account for the relationships among several spaces, which can drastically affect users' movement and activities.

Virtual reality studies are commonly used to overcome the costs and logistic challenges of constructing physical mock-ups. Multiple users can inhabit virtual spaces and test to what extent a building design affects users' movement and activities [22]. These kinds of studies, however, for technical and operational reasons are usually constrained to a limited number of users that can participate in the study. Furthermore, a poor design of virtual reality can potentially hinder realistic response from the users [23].

Static analyses of human behavior analyze the geometric and topological properties of a building layout to infer building-human interactions. Space Syntax, for instance, is a well-known mathematicallyinclined method for studying the impact a built environment on users' cognition and navigation [24–28]. Spatial models have also been integrated with descriptions of people activities for space-use analysis [29–35]. However, these aforementioned approaches fail to grasp the dynamic, time-based nature of human behavior especially in complex settings, where multiple agents with different roles engage in individual or collaborative activities that impact (and are impacted by) a spatial and social context.

Data-driven approaches predict users' presence and actions in openplanned spaces [36] and mostly in office buildings to analyze users' comfort and energy consumption [1,37,38]. These methods, however, can be used when large occupancy data sets are available – which is not the case of not-yet-built environments. Additionally, they often ignore the impact that a building design produces on dynamic aspects of human behavior.

2.2. Simulating human behavior in built environments

Simulation approaches have been proposed to analyze the dynamic relationship between human activities and the surrounding environments in both existing and not-yet-built environments. Particle-based methods describe pedestrians as homogenous particles subject to physical and social forces of attraction and repulsion [39]. Fluid-based methods describe people flow in fluid-like terms [40–43]. Cellular automata models provide an inherently spatial representation of occupancy, whereby each cell indicates its occupancy state and transition rules govern the evolution of a cell state [44,45]. Process-driven models consider structured sequences of activities that require a set of resources (e.g., people, equipment) and take a certain (usually stochastic) amount of time [46]. In these models, space is often abstracted in the form of a graph where nodes represent rooms and link represent stochastic traversal times [47]. In hospitals, however, several processes may take place in the same space, and one process may affect the other. The aforementioned approaches, however, cannot consider interactions among multiple parallel processes occurring in the same space. Unplanned social interactions between staff members and patients, for example, have been proven to affect the performing of other medical tasks [20].

Different from these approaches, in Multi-agent systems (MAS), autonomous agents inhabit virtual environments and sense, plan and act individually or in groups to achieve a specific goal [48,49]. Several approaches focus on simulating pedestrian movement in normal [9,50] or emergency scenarios [10,51–54]. These methods consider linear and straightforward behaviors (e.g. move towards a goal), which are not representative of day-to-day human behavior in complex facilities.

Research on MAS that mostly stems from computer graphics focused on modeling crowd behaviors aimed at producing realistic simulations of Non-Player Characters (NPC) in video games or conducting safety assessments in evacuation scenarios [11,55]. While these approaches provide efficient solutions to simulate collision-free movement and social interactions [56], they mostly focus on the abstract movement of occupants while ignoring the setting where a behavior is enacted (e.g. a specific building type) and the context-dependent activities that people engage in (e.g. task-based behaviors in healthcare facilities).

Other approaches proposed pre-occupancy evaluation methods to predict neighborhood safety [57] and day-to-day scheduled activities in university buildings to improve designer-user communication [14,58]. The first approach, however, considers reactive behaviors of single virtual users in response to spatial properties of a built environment (e.g. the presence of a balcony or street lights). The second approach, instead, generates fixed schedules that cannot dynamically adapt to unplanned activities, such as impromptu agents' interactions that are enabled by the proximity of the agents in a given space, at a given time. 2.3. Narrative-based modeling

Recent work on narrative-based modeling [16,59] demonstrated a different approach to simulating day-to-day occupancy scenarios in complex facilities, like hospitals. The approach is centered on narratives, rule-based scripts that coordinate the collaborative behaviors of heterogeneous actors (e.g. doctors, nurses, patients) to perform a structured sequence of activities (e.g. checking a patient) that unfold in semantically rich spaces (e.g. patient rooms). Different from other multiagent systems where agents are either equipped with autonomous decision-making abilities (e.g. [57]) or are directed by a centralized scheduling method (e.g. [14]), the narrative-based model uses a combination of a top-down coordination mechanism to enforces the performing of structured sets of tasks, while allowing for bottom-up adaptations to dynamic social and spatial conditions, such as the emergence of unplanned narratives (e.g. staff-visitors interactions) that can potentially delay planned narratives (e.g. checking a patient).

A key aspect of this approach involves distributing ‘intelligence’ among the different components of the model. Narratives, in fact, are responsible for the high-level organization of atomic activities into taskbased procedures. To relive the activities and narratives from the need to handle low-level calculation processes, both actors and spaces entities are equipped with autonomous calculation abilities to dynamically update their status based on contextual social and spatial conditions. Such status can be retrieved by activities and narratives during their execution so that they can make the most informed decision at any given time.

narrative manager determines which narrative to trigger at a given time, depending on the current state of the world (e.g., the current simulated time or the proximity of actors in a space). In contrast to other

https://lh3.googleusercontent.com/dgxYPu6j7lZ7tJLA5XAJ5SjsZqIVR_eZXrMm-jzkBS92z150KNXBN8lTQyLLuIMPFgDwM1v6tJgh6IV7KNkhGyaXH29HlLvmqpnhA5bCNCCs9ofvn8xlRH4GvcubPp5wFzVPfOk

Fig. 1. Narrative-based model architecture [16]. A narrative manager monitors the states of the world and triggers planned and unplanned narratives when specific preconditions are met. When triggered, a narrative directs the behavior of one or more agents by associating them to a specific activity.

approaches that simulate scheduled activities in workplaces, the execution of narratives can adapt to dynamic conditions. For instance, unplanned narratives (e.g., staff-patient interactions) can cause delays to planned narratives (e.g., a patient check). Fig. 1 provides an overview of the narrative-based model.

While a similar definition of narratives can be found in research on digital storytelling and video games, where narrative entities coordinate the unfolding of overarching stories involving the collaborative behavior of Non-Player Character (NPC) agents [11,60], these approaches do not adopt a ‘distributed intelligence’ approach and they fail to account for the impact that dynamic social and spatial conditions and unplanned narratives produce on the story unfolding.

Prior work demonstrated the ability of the narrative-based model to account for representative collaborative planned and unplanned narratives in abstracted hospital environments. However, the scalability of this approach to model human behavior patterns in real-world hospital settings and its ability to calculate key performance indicators of building occupancy that can be visualized and analyze by design stakeholders has not yet been demonstrated.

To address this issue, we developed a narrative-based pre-occupancy evaluation platform that extends prior capabilities of the narrative-based models with new features required to simulate human behavior patterns in real-world hospital design projects and analyze and evaluate key performance indicators on how buildings may operate when built and occupied. We then applied the platform to conduct the pre-occupancy evaluation of two alternative design options for an outpatient ophthalmology clinic. We detail the design of the proposed platform in the following section.

3. Narrative-based pre-occupancy evaluation platform

To examine the hypothesized scalability and applicability of the narrative-based model to conduct comparative pre-occupancy evaluation of design options for real-world outpatient clinics, we developed a platform, which involves the following components: (a) a library of spacesactorsactivities and narratives that represent respectively the spaces that people inhabit, the actors that populate the spaces, the activities they perform, and the narratives they are involved in (b) a narrative manager that coordinates the narrative unfolding (c) a simulation engine that calculates the behavior of the entities over time and (d) and a pre-occupancy evaluation module that calculates and visualizes a list of KPIs so that design stakeholders can compare the results with their goals and expectations. In this work, we use Unity 3D as simulation engine. Unity 3D is a popular video game engine that features advanced physics and artificial intelligence libraries to model collision avoidance and path-finding. Fig. 2 outlines the different components of

https://lh6.googleusercontent.com/c3PAOH9MVPR2RZx9b9hqECUZOOt9lSkndYG4xRVPH0_gOzN39FFtkbN4ukqfRd9DDkYD3n236qaLK3KDKiaaJ-p6aZW0nFRMtXfjC3Y2c5-uWkSGRhxqb1EXlg8zWXUkZ5xq-U0

Fig. 2. Narrative-based Pre-Occupancy Evaluation Platform. Design stakeholders inform the modeling of space, actor, activity and narrative libraries (highlighted in orange), which are used as simulation input. The simulation results are analyzed with respect to a selected number of Key Performance Indicators (highlighted in cyan) and evaluated in collaboration with different design stakeholders. (In the narratives' library, N-D = Nurse-Doctor, and N-P = Nurse-Patient). (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

the platform and its use in collaboration with design stakeholders.

The platform extends prior work on narrative-based modeling by introducing novel space, actors, activities and narrative libraries capable to represent human behavior in real-world outpatient clinics. It also introduces a newly implemented evaluation module that calculates and visualizes KPIs tailored to hospital settings (e.g., nurse-visitor unplanned interactions). The analysis of the selected KPIs is made possible by a combination of pre-existing narrative-based model features (e.g. the model's ability to simulate the dynamic interactions between planned and unplanned behaviors) and newly implemented features embedded in the entities' libraries (e.g. the narrative's ability to dynamically account for the impact that a building design produces on their unfolding). We detail each component of the platform in the following sections.

3.1. Modeling

To simulate building-human interactions in real-world outpatient clinics, it is first necessary to collect data about human behavior patterns that are expected to take place in the designed environment and model them using a library of modular and reusable space, actor, activity and narrative entities, able to coordinate the collaborative behavior of building occupants, while accounting for the impact that a building design produces on the unfolding of planned and unplanned narratives.

3.1.1. Data collection

Semi-structured interviews were conducted with clinic managers, head nurses and administrative staff in outpatient ophthalmology clinics in Israel to inform the modeling phase. Data has been gathered about the different types of occupants, the different functions of the spaces they inhabit, the types of individual and collaborative activities they engage in, and the task-based procedures they undergo. Although the patients that visit ophthalmology clinics may suffer from different pathologies, the activity flow of such patients can be considered similar at the level of abstraction used in this study. An example of the flow of regular patients is shown in Fig. 3. A different flow is followed by Emergency Department (ED) patients, who are admitted for a doctor's treatment without checking in and out at the reception desk. Typical patients are usually accompanied by (at least) one companion. Physicians are expected to spend most of their time in their offices to treat patients, while nurses are in charge of performing visual acuity tests and perform other tasks that require frequent transitions between rooms, meeting with the doctors, patients, and their companions.

https://lh4.googleusercontent.com/0CtXlMwxW_OsriQB9s6ib2nX7Y2ACvin4-Ev3JmQ72u55HjJw3XtOX8dQeMKS4gcLF3r3WB2kJfedsm6UHDy5OQlAK8wQqu5pQ-uSxDIi70bd_wFzZNpOKx89CvPDmxywwmpKTM

Fig. 3. Activity flow of regular patients followed by their companions (P = Patient, C = Companion, S = Secretary, N = Nurse, D = Doctor).

Based on field observations in existing spaces, discussions with the head nurse, and previous findings [61], we have identified behavior rules that characterize the impact that a built environment produces on human behavior. For instance, we stipulated that patients and their companions wait as close as possible to their destination target when waiting (e.g., the doctors' or nurses' rooms). Furthermore, in addition to the considered “planned” activities, we have also considered “unplanned” staff-patient interactions that delay the performing of planned activities [20,62,63]. These abovementioned insights informed the modeling of multi-agent narratives and their components.

3.1.2. Space models

Space models are composed of two different components. The first is a building model generated using CAD or BIM tools, which comprises both physical and non-physical components. Physical components include walls, floors, doors, furniture, seats, etc. Non-physical components include rooms, corridors, and open areas. We model these latter components in the form of zones, discrete portions of space that host specific activities, such as treating a patient or waiting [64]. Both physical and non-physical building components, beyond geometric information, store semantic labels that indicate how they can be used. A ‘clinic room’ zone, for instance, indicates that the space can be used for clinical activities, such as treating a patient. To simulate human behavior narratives, we augment space models with additional data, such as spots, specific target locations for agents' activities that are not directly specified in the geometric model, which indicate, for instance, the locations for standing in a queue or in a waiting area.

The second component is modular and reusable profiles, which provide zones and spots with static properties and dynamic calculation abilities. Zones, for instance, can record the presence and activities of the occupants within their boundaries. To represent building-occupant interactions in real-world hospital settings, we developed a novel library of zone profiles that empower space with dynamic calculation abilities aimed at supporting the narrative performance. Namely, we consider three types of zones: ‘waiting’, ‘queuing’, and ‘clinic’.

  • Waiting zones store and update a list of waiting spots within their boundaries as well as their dynamic status (e.g., available or occupied). Waiting spots can be ‘sitting spots’ or ‘standing spots’. To identify a place to wait, an activity can communicate to the zone a list of group members and a final target. Based on the current spot availability and the proximity to the target destination, the zone identifies a list of adjacent spots where agents can wait while prioritizing sitting spots adjacent to the target destination, as identified in the data collection phase.
  • Queuing zones store a list of queuing spots where agents can queue while waiting at a reception desk. Queuing spots have a status indicating their availability and the ID of the actor occupying it. The zone dynamically monitors the availability of queuing spots and communicates a list of available spots when queried by an activity or a narrative.
  • Clinic zones are associated with specific doctors or nurses. They record the presence of staff members in the room and keep track of the patients waiting to be treated by the staff located in the room. Every patient waiting for a doctor or a nurse notifies the clinic zone, which, in turn, records the patient ID. Every time a nurse or doctor completes a treatment, the zone automatically updates this list and notifies the new patient to be treated. If a nurse exists his/her room, the room changes its status, so that no patient is allowed to enter. Once the nurse reaches their desk, the room will update its status accordingly to allow patients in.

Fig. 4 provides an example of a space model. We summarize key properties of space models in the Appendix A (Table A1).

3.1.3. Actor models

Actor models include shape properties, which can be abstract, as it is done in this study, or more detailed as in Shen et al. [14], and a profile (e.g., patient, companion, doctor) that determines the type of narratives the actor can be associated with. Additionally, the profile equips actors with static and dynamic properties. Static properties store information about associated users or spaces. For instance, each patient involved in a “patient treatment” narrative can be associated with a companion as well as a doctor and nurse responsible for their treatments. Nurse users, instead, can be associated with a number of patients to treat as well as an office space. Dynamic properties store information about the current activity they are performing, their current waiting spot, their queuing position to visit a doctor, etc. Much like spaces, the information calculated and stored in actor models can be used by activities and narratives to drive their behavior [65]. The library of actor profiles developed in this study consists of ‘patients’, ‘companions’, ‘nurses’, ‘doctors’, and ‘secretaries’.

• ‘Patients’ undergo a medical treatment while being assisted by ‘companions’. Accordingly, patients and companions are combined into groups. Each patient can be associated with a different number of companions. However, only one companion can enter with the patient in the examination rooms; the others, are instructed to wait in the waiting area. The group size affects human spatial behavior in that, in this work, we stipulated that group of agents will seek places to wait for medical treatment in zones where all group members can sit or stand. Accordingly, the group may wait far from a target destination if no available spots can be found in closer zones. This will affect the group walking distance, the overall completion time of the narrative, and the chances for social interactions. Both patients and companions store dynamic information with respect to their current sitting and waiting locations. Different from companions, patients store and update information related to the doctors and nurse they will visit (e.g. their waiting position in the queue) and the number and location of interactions with the staff members. • ‘Secretaries’, ‘nurses’ and ‘doctors’ represent different types of staff members. Secretaries are static in the reception area and they perform administrative duties for patients who queue at their desks. Nurses and doctors are associated with a specific ‘clinic’ space and a target spot that indicates their desk location. Different from doctors, nurses participate both in planned and unplanned narratives, the latter of which require them to exit their office to interact with doctors or patients. Each nurse stores in their profile the number of time he/she exited the office. This information is used by the narrative manager to determine when the nurse can participate in this narrative.

Fig. 5 shows an example of an actor modeled in Unity, with an

https://lh6.googleusercontent.com/K2nBcrqPaRV2Dij-DcWlrX0XPkcAYPDdbXcCLfAb1AV_fGQZdgyMMiwmC_EwS61fkTWZ5bG1kOj7NtDrYjGlPN9rdMoP2wGWSAluqO6bsp_8We1XwtwySs6EIB2U9LCIAHZP57w

Fig. 4. Example of a space model with related properties. In red, we highlight a specific zone with its associated profile that stores static properties (e.g. zone semantics) and updates dynamic properties (e.g. number and ID of the agents within its boundaries). (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

https://lh5.googleusercontent.com/bSHLJsIBX4NfxsEaJaStizHbZlu-nQZN5YsbYCB_mmb9OuZSq84IOSp8P3N0TQVEBW5BI1f43y_ZJ-BBi9_fYDcV77XOS29E-VabHxwoRBOh4qJO53u1UdiA-Z2rSMveouclYkw

Fig. 5. Example of an actor model with a related profile.

associated profile. We summarize key properties of actor models in the other users or with the built environment. In this work, we are con-

Appendix A (Table A2). cerned only with an abstracted description of the activity (such as a medical treatment), its spatial location, the identities of the participating actors, and duration of the activity. This way we can limit the

3.1.4. Activity models number of activities modeled and focus on their spatial/social

Activities represent the possible interactions that users have with

implications in real-world outpatient clinics. Specifically, we identify four activity patterns that can describe building-user interactions in real-world hospital settings: they are moveinteractwait and queue.

  • Move controls one or more actors moving to a target. It uses the A* algorithm to direct a group of agents towards a target destination using the shortest available route. The activity is completed when all the actors reached the specified target.
  • Interact coordinates the interactions between a group of actors for a specific time duration. Depending on the actors participating in the interaction, it can represent a conversation between a patient and a secretary as well as a patient check activity involving a patient and a doctor. The activity is completed when the specified activity duration is over.
  • Wait involves a list of actors waiting in a specific waiting spot for a condition to occur. For example, a patient and companion wait for their turn to be received by a doctor or a nurse. Alternatively, a patient can wait for a specific duration of time for their eyes to dilate. Narratives are responsible for specifying the type of end conditions (i.e. either time or the end of a turn to be checked by a medical staff) and for defining a specific spot where actors can wait using information provided by ‘waiting’ zones (as specified in Section 3.1.2).
  • Queue handles the queuing process of patients and companions waiting in a line to interact with a staff member (e.g., a secretary). After identifying an available queuing spot by means of a ‘queuing zone’ (as described in Section 3.1.2), an activity directs the actors to the spot and instructs them to wait until the next available queuing spots are available. The activity is completed when the actors reach the end of the queue. In this case, the distributed intelligence approach supported by the narrative-based model significantly contributes to simplifying the simulation of a queuing process. In fact, such a calculation is split between spaces, actors, and activities: the space calculates the first available spots and indicates the ID of the adjacent waiting spots, the actor stores the ID of the current spot where the agent is located, and the activity monitors the status of the next available spot and instructs the agent to move to the next spot when it becomes available.

Such activity patterns, depending on the actors involved and the space where they take place, can describe various behaviors observed in hospital wards. We model each of these activities in a modular fashion, so they can be reused multiple times within a narrative or across narratives to describe context-specific behaviors in real-world settings. For instance, the same interact activity can be used to describe a visual acuity test – if associated with a patient, companion, nurse, and a clinic room – or a doctor treatment – if associated with a patient, companion, doctor, and a doctor's room. We summarize key properties of activity models in the Appendix A (Table A3).

3.1.5. Narrative models

Narratives are the heart of the simulation. They use the aforementioned components (spaces, actors, and activities) and combine them into scripts that direct actors' behavior by associating them with specific activities performed at a given time and space. Narratives include 3 main components: preconditionsperforming steps, and postconditions [16]. The preconditions define a specific state of the world (e.g., time or actors' and spaces' dynamic status), which must be satisfied for the narrative to be triggered. The performing steps consist of a structured sequence of activities associated with specific parameters calculated by the narrative entity. The postconditions determine the changes that occurred in the state of the world due to the performed narrative.

Different from previous approaches that adopted narrative-based modeling, in this work we empower narratives with additional calculation abilities beyond storing a sequence of activities to perform. They also contain rules to calculate relevant context-dependent parameters required for performing the specified actions. In fact, while the narrative manager assigns to a narrative a specific set of inputs (e.g., the list of actors involved in the narrative), other input values cannot be precomputed since they depend on the dynamic state of the world at the time of activating the narrative. For example, a narrative can assign a group of actors specific waiting spots that will be as close as possible to the target location (e.g., a doctor's room), while accounting for the dynamic availability of waiting spots in each zone. The narrative can calculate such information by querying ‘queuing zones’, which store dynamic information of currently available waiting spots, and pass it as an input parameter to the waiting activity that will direct the actor to sit in that specific spot.

This newly implemented library is composed of four types of narratives, namely ‘Regular Patient Flow’, ‘ER Patient Flow’, ‘Nurse-Doctor Interactions’, and ‘Staff-Patient Interactions’. The first two narratives are planned since they take place when the patient and companion enter the ward. The other narratives are unplanned since they can be triggered anytime during the simulation, when their preconditions are met.

  • The Regular Patient Flow narrative directs both a patient and a companion that just entered the ward to register with a secretary at the reception desk, meet a nurse for a visual acuity test, wait for the patient's eyes to be dilated, wait for the doctor, meet the doctor for treatment, check out at the reception desk and then leave. Upon being triggered, the narrative manager associates with the narrative a list of actors including the patient, companion, doctor and nurse associated with the patients. Other parameters, such as the queuing spots and the waiting spots of the actors, are dynamically calculated by the narrative during its execution.
  • The ER Patient Flow narrative directs an ER patient and a companion to wait for a doctor in specific ER-dedicated waiting spots, meet a doctor for treatment, and then leave the ward.
  • The Nurse-Doctor Interaction narrative coordinates the behavior of a nurse who leaves the office after visiting a determined number of patients to assist a randomly selected doctor. After interacting with the doctor, the nurse goes back to his/her office.
  • The Staff-Patient Interaction narrative takes place when a patient and nurse cross paths in the corridor. Specifically, we have considered such interactions to take place every time a nurse meets a patient or its companion waiting in the waiting area to be treated by either a nurse or a doctor. We have also stipulated that nurses and patients/ companions, after engaging in a social interaction cannot further take part in such a social interaction narrative for a short time period, arbitrarily set as 2 min.

Fig. 6 provides an overview of the narratives considered in this study. We summarize key properties of activity models in the Appendix A (Tables A4, A5, A6, and A7).

3.1.6. Narrative manager

https://lh4.googleusercontent.com/3TIVqn9NE_zClqXQ7v_NM1vu7_S0tv0UqyWkrXlEQiB9oy2ax5a__oq8G9-RLtE72u2YYfrsieLZfAYahBPvVpcjh0BQdakcJlB_aK-TGb_j2W-jsqgwqQBkBCtDJz9pNQiTudo

Fig. 6. Narrative flow charts.

https://lh5.googleusercontent.com/nZfZPWmKsInhBLU-LjkpDI3cPd2sEr2gR3y6_hn4FOwJPe99AToT9-pXa_hstDUoyzoYIqkTaHVSXAtAQF14qhDU_nwtVw5iJYQVN--LAZ0J_dXXmRuUDRuOh0rvfjhEAfKs4O0

Fig. 7. Narrative Manager algorithm. It spawns actors and it coordinates the performing of planned and unplanned narratives.

The narrative manager coordinates the narratives' unfolding over time. Different from prior work, we empowered the narrative manager with the ability to instantiate groups of actors into the clinic ward and associating them with a planned narrative to participate in (either regular or ER patient flow). During the execution of planned narratives, the narrative manager is also responsible for triggering unplanned narratives when the necessary preconditions are satisfied. To prevent the narrative manager from constantly monitoring the state of the world (e.g., actors' statuses or proximity in space), we have equipped actors and zones with the ability to send a message to the narrative manager when the conditions for triggering an unplanned narrative are met (e.g., the proximity of a nurse and patient, or a specific number of patients has been checked by a nurse). Based on the received information, the narrative manager temporarily suspends the current narrative the agents are involved in and triggers the unplanned narrative. Upon completion of the unplanned narrative, the unfinished previous narrative is resumed. When the actor completed its planned narrative, the narrative manager removes the actor from the ward while storing its data in a persistent database. Fig. 7 summarizes the key steps of the narrative manager algorithm.

3.1.7. Pre-occupancy evaluation

The simulation results are evaluated with respect to multi-criteria KPIs defined in collaboration with the different design stakeholders. This platform supports the calculation and visualization of the following KPIs: (a) patients and nurses' walking paths and distances, (b) patients' length of stay and overall throughput, (c) people density over time, and (d) number and location of social interactions. These metrics have been selected based on insights gained through interviews with staff members of existing ophthalmology clinics as well as published literature. Specifically, long travel distances can cause delays in medical procedures and can cause staff and patient fatigue [63]; increased people density can produce spatial congestion, which can limit the flow of people and equipment and produce high noise levels [62]; an

Fig. 8. Graphic User Interface that enables real-time analysis of occupants' spatial behavior. (A) shows a simulation clock, the number of patients that are in the ward and the ones who left. (B) shows real-time spatial analysis such as actors' walking paths, distances, time of permanence in the ward, and spatial density. (1) and (2)

provide an example of data that can be visualized.https://lh3.googleusercontent.com/lGkjQR_NlWgHQBPRhIzPbgkw00O74wsqj8WFR3rwX6EPiGdp26Dp6blm2pepbWQDOLqHNpx4K6G8qgoDUn8nI_jhExZ7XIpNXRFZMWcw4zGooodCj7Qxkpc-olhh1-B2YC3sH80

elevated number of staff-patient interactions can be considered beneficial for patients, but disruptive for caregiving procedure, since they can lead to spatial congestion, delays, and both medical and administrative errors [20,66].

Such KPIs can be visualized and evaluated in real-time during the simulation by means of a graphic user interface (GUI), which displays the dynamic activities of actors in a proposed building layout (Fig. 8). Additionally, the GUI represents metadata including a clock that indicates the current hour in the simulation, and the number of patients and companions that are still in the ward and those who have already left. A control panel enables the user to visualize real-time numerical results, such as the distances walked by the patients, and data-maps – spatio-temporal representations of occupancy-related data over time, such as the walking paths of patients and nurses [67]. Numerical values generated by the simulation are stored in a database. Aggregated data values are also available, including numerical KPIs such as a patient's length of stay in the ward, people density over time in given spaces, and the aggregated paths of different user types (i.e. patients and nurses).

Both numerical data and the data-maps are discussed with the different stakeholders who can compare the obtained results with their goals and expectations. Additionally, the stakeholders can interpret the simulation results by revealing implications that are not self-evident directly from the data produced by the simulation. For instance, while the simulation records the paths of nurses and patients, the potential implications of spatial bottlenecks in the different areas of the hospital ward (e.g., potential interference with doctors' work due to increased noise and potential interruptions) are not directly evident from the simulation. They are extrapolated by means of a shared discussion with the stakeholders, who can evaluate the results while relying on their knowledge and experience.

3.2. Implementation details

Spaces have been modeled using Autodesk Autocad and then imported into Unity 3D using the FBX format. Actors have been modeled directly in Unity as capsule-shaped objects. The spaces, actors, activities, narratives and narrative manager have been modeled directly in Unity 3D using Microsoft C#. Activities and narrative have been modeled as co-routines. Narrative co-routines are composed of a structured set of activity coroutines that are nested within the narrative

Fig. 9. Two architectural design models for an ophthalmology outpatient clinic. In Design A, the waiting area is centralized, while in Design B it is distributed.We have masked in white the parts of the ward that have not been considered in the study.

5Credits: Mochly-Eldar Architects for Design A and architect Faten Tabry Kattouf for Design B. The original floor plans provided by the architects have been

abstracted by the authors to conduct this study.https://lh6.googleusercontent.com/uBYEeIvY9A0LEJ6hNJtAZ59-3o-uVWSBKR2qNKChNOxf2oIxeuQed3tQyCfHwukBnNpnDNo8dfqH_bSNZof5CrdKSHoo6zxH4dxtxbcCygVtPa1VKjcZNzyVPzYwarXzYIwJ1P4

and executed one after the other while yielding at each time step a status to the parent narrative (e.g., running or completed). Based on such a status, each narrative updates its own status, which is reported to the narrative manager. To run multiple narratives involving different actors concurrently, we have leveraged Unity 3D's ability to run coroutines in parallel, emulating multi-threading processing.

4. Case study

We demonstrate the proposed framework by means of a case study, which compares the performance of two different architectural designs for an ophthalmology outpatient clinic. Ophthalmology clinics host large volumes of patients and visitors that must go through a series of procedures to diagnose and treat eye-related issues. This study has been conducted in collaboration with two major hospitals in Israel: the Rambam Health Care Campus in Haifa, and the Meir Medical Center in Kfar Saba. Both hospitals plan to expand their ophthalmology outpatient clinics to improve patient flow and optimize operational efficiency. Each hospital produced a new design for its ophthalmology clinic. In this study, we compared the performance of these different architectural layouts to analyze how the different building designs may affect day-to-day occupancy patterns.

The main differences between the two designs concern the organization of the waiting areas for patients and visitors (Fig. 9). In the case of the Rambam Medical Care Center (Design A), the waiting area is centralized and is located near the entrance of the ward. A locked door prevents patients and companions from accessing the area where the doctors' offices are located unless it is their turn to be received. In the case of the Meir Medical Center (Design B) the waiting areas are distributed near the ward's entrance and the doctors' treatment rooms. In this design option, patients are allowed to wait close to the doctors' offices. These space layouts reflect two different design goals of the stakeholders (hospital managers, end-users and architects). In Design A, the goal is to maintain the treatment area free from patients waiting in the corridors to prevent staff-visitor interactions and potential disruptions of treatment procedures. In Design B, the goal is to locate waiting areas as close as possible to the treatment areas to reduce patients' walking distance and to increase staff oversight of the patients.

Beyond such differences in space organization, the two design alternatives also differ in size (1240 m2/13,240 sq. ft. for Design A, and 860 m2/9260 sq. ft. for Design B); children treatment strategies (in Design B, the children clinic is separated from the ward); patient flow expectancy (Design B is equipped for treating more patients); and number of staff members (Design B is expected to employ a larger number of staff members). To isolate and focus on the implications of spatial factors on occupancy-related metrics, we have considered an equal functional area of 500 m(5380 sq. ft.) in both clinics and we have simulated the same activities and the same number of users in both layouts based on projections of future building operations provided by the hospital managers. We detail each step of our pre-occupancy evaluation study in the following sections.

4.1. Simulation setup

The space models of the two ophthalmology clinic designs considered in this study have been previously shown in Fig. 9. Each space model is comprised of a series of zones with different profiles, such as ‘clinic’, ‘waiting’, and ‘queuing’ (as described in Section 3.1.2). Spaces are populated by actors with different roles, such as ‘doctors’, ‘nurses’, ‘patients’, ‘companions’, and ‘secretaries’ (as described in Section 3.1.3). Actors are engaged in four types of activities, namely ‘move’, ‘meet’, ‘wait’ and ‘queue’ (as described in Section 3.1.4). Activities are organized in four types of narratives, namely, ‘regular patient flow’, ‘ER patient flow’, ‘nurse-doctor interactions’, and ‘staff-patient interactions’ (as described in Section 3.1.5).

In this study, we considered 150 patients, which is the number of

Table 1

Simulation input defined in collaboration with the design stakeholders.

Type

Parameter

Value

Actors' numbers

Patients Regular

140

 

ED

10

 

Doctors

10

 

Nurses

2

 

Secretaries

2

 

Companions per patient

1

Activities' duration

Talking with secretary

2.5 min

 

Visual acuity

7 min

 

Eye dilation

30 min

 

Patient check

20 min

 

Nurse-doctor interaction

2 min

 

Staff-patient interaction

30 s

https://lh4.googleusercontent.com/sbZDEXWVj8ILvmQgOpxy1BFuN7OeC9K2LOO4TZ03oFSLLwaQ_m0eGCCgQS-IRnQ0nm5jzOrql0BTwcQvltA6inZR-cV0hlCqlmfc7Bd2AEX9R_hAfl-3iXjeUBJULo4FO1IEKsI

Fig. 10. Simulation snapshots at peak hours (11 am). Blue dots represent patients, green dots represent companions, orange dots represent nurses, red dots represent doctors, and pink dots represent secretaries. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

Table 2

Simulation results.

Key performance indicators (KPI)

   

Design A

Design B

Travel paths and distances

Paths

Patients

 

Fig. 11

 
  

Nurses

 

Fig. 12

 
 

Distances

Patients

Max

568 m

197 m

   

Average

146 m

117 m

  

Nurses

Max

2620 m

3069 m

   

Average

2399 m

2548 m

People density

   

Fig. 13

 

Staff-patient interactions

Locations

  

Fig. 14

 
 

Number

  

93 interactions

119 interactions

Patients' throughput

Patients' length of stay

Max

 

4 h 22 min

5 h 07 min

  

Average

 

3 h 1 min

3 h 15 min

 

Overall patients' throughput

  

9 h e 5 min

9 h e 39 min

patients expected to be treated in the future clinics (according to the interviewed staff). Among them, we considered 140 patients are regular patients, while 10 are ER patients. Additionally, we considered 10 doctors, 2 nurses and 2 secretaries. While the expected number of doctors and nurses working in the clinic is slightly different between the two hospitals, we arbitrarily chose to use the same number of users in both scenarios to isolate the impact that the space will produce on building occupancy factors. Additionally, we assigned similar representative durations for the considered medical and administrative activities (as indicated in Table 1).

During the simulation, patients are spawned into the ward in regular ‘pulses’, every 10 min, starting at 7 am (the beginning of the shift). At each pulse, 10 patient-companion groups are spawned. Among them, 9 patients are regular patients, while one of them is an ED patient. The narrative manager algorithm (previously described in Section 3.1.6) resolves conflicts among narratives competing for the same agents. In

https://lh3.googleusercontent.com/uqCwYvAteSUSG7B1tsizpp5zXnRXKjEY3fsn9p5D9X5UVJuRT7WyxTIwoq_fF7kT-64Z_pe1wc3NLv5du5TdU16G_ggCnHTGWQa6lI-HmkUuwbsB7tUSCAcmo8zu8S82Tb7o9lc

Fig. 11. Aggregated patients' walking paths (colored in light blue). (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

https://lh5.googleusercontent.com/2RStUJnnRCd1RcFoOOQVegnU8sU9GblYdYvRrl51ZJF06UClbvCB0OrZRs41QFyEWYBaGJ26Sw_Z21rwZqjJcm1fvDpPuvuD0BdaVBd7n5PRvP1blhj8NU10xBNHK94CNFqOHh4

Fig. 12. Aggregated nurses walking paths (colored in orange).

https://lh3.googleusercontent.com/xi1LT8r9Aij2Hq_ouE_fhcK5pRyD_3PlofJ8vIER8axN_CQu6nR9qPYeC9-ngQhH_2T89esKFvtbcdG52NKCJTAZg73RGP-IVsovK-9u3Zy-MyTsC4u4pUmWjEVtMIN7ogDMwjQ

Fig. 13. Aggregated people density. Redder zones represent denser areas.

https://lh3.googleusercontent.com/U1ICLd4-wx1CF-TIkjJqe_60J22p9QXvhdrSyyNmnv7BJLVmOK8kCyig1JU2Su8LLkbkp_QObrWcUKEeM2PO93jhK4GVjHJ5gYpOOjJbqYJeLJs0MUEALPd8r0uIb1nWnr0usqQ

Fig. 14. Aggregated location of staff-patient interactions (indicated as yellow dots). (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.) this study, we have stipulated that, when their preconditions are sa- 4.2. Simulating multi-agent narratives and evaluating the results tisfied, unplanned narratives are always triggered. Table 1 summarizes

the simulation input parameters. Fig. 10 shows a simulation snapshot at 11 am for each one of the considered designs. That time is considered to be the peak of people

Table 3

Pre-occupancy comparative evaluation of two architectural design options, while considering the stakeholders' design goals, and potential implications of the design solution to affect users' satisfaction and operational performance.The green color indicates that the design options provided a better match with the design goal compared with the other design alternative. The yellow color indicates the opposite.

Key performance indicators (KPI)

Goal

Design A

Design B

Result & implications

Result & implications

Travel paths and distances

Paths

Patients

Close to Doctors' rooms

Reduced interference with staff operations; improved efficiency of doctors.

Increased interference with staff operations; diminished efficiency of doctors.

Close to Nurses' rooms

Increased interference with staff operations; diminished efficiency of nurses.

Reduced interference with staff operations; improved efficiency of nurses.

Nurse, in front of waiting area and reception

Increased changes of social interactions and interferences. 

Increased changes of social interactions and interferences. 

Distances

Patients

Diminished patients' 

satisfaction and possible delays if the patient arrives late to the visit.

Increased patients' satisfaction and operational efficiency.

Nurses

Reduced tiredness; increased satisfaction and operational efficiency; increased time spent with patients; reduced chances of mistakes.

Increased tiredness; reduced satisfaction and operational efficiency; reduced time spent with patients; increased chances of mistakes

People density

Doctors' rooms

Reduced noise and interference with staff operations; improved efficiency of doctors. 

Increased noise and interference with staff operations; diminished efficiency of doctors.

Nurses' rooms

Increased noise and interference with staff operations; diminished efficiency of nurses.

Reduced noise and interference with staff operations; improved efficiency of nurses.

Staff-patient interactions

Number 

Reduced interruptions; improved efficiency; reduced chances of mistakes

Increased interruptions; reduced efficiency; increased chances of mistakes

Locations

Doctors' rooms

Reduced interference with staff operations; improved efficiency of doctors

Increased interference with staff operations; diminished efficiency of doctors.

Nurses' rooms

Increased interference with staff operations; diminished efficiency of nurses.

Reduced interference with staff operations; improved efficiency of nurses.

Patients' throughput

Patients' Length of stay

Improved patients' satisfaction and overall operational efficiency.

Improved patients' satisfaction and overall operational efficiency.

Overall throughput

Improved operational efficiency.

Diminished operational efficiency.

We acknowledge that this table is not comprehensive and that some criteria considered in this study may have mixed effects on the building operation (i.e. both

beneficial for some aspects and disadvantageous for other aspects).

density in the ward. Groups of actors are queuing at the reception desk, (which we set to end when the last patient left the clinic), we evaluated waiting in the waiting area, and undergoing treatment from nurses and a number of KPIs identified in collaboration with the stakeholders doctors. (nurses, doctors, and hospital managers) with respect to their goals and

After running the simulation for the complete duration of the shift expectations. Specifically, we considered the following KPIs: (a)

Table 4

Comparison with prior work on narrative-based modeling.

Feature type

New features compared with Schaumann et al. [16]

Entities' Library Spaces Actors

Activities

Narratives

Narrative manager

Pre-occupancy evaluation

Case study

  • New ‘waiting’, ‘queuing’, and ‘clinic’ profiles that account for the dynamic impact of space on people while waiting for treatment or queuing atthe secretary's desk.
  • New ‘patient’, and ‘companion’ profiles, which can be associated into groups and can participate in newly modeled waiting and queuingactivities.

••• New features added to ‘nurse’ profiles, who can engage in unplanned narrative to interact with doctors.New ‘waiting’ and ‘queuing’ activities to manage human spatial behavior.Improved modularity and reusability in order to compose complex narratives using the minimum number of activity types (4 types of activitieare used to simulate 26 types of behaviors organized in 4 narratives) s

•• New features to dynamically calculate parameters and communicate them to activities to account for dynamic spatial and social features.New ability to spawn agents at specific time steps, associate them with planned narratives, and remove agents from the ward when narrativesare completed.

  • New calculation of key performance indicators, which include: (a) travel paths and distances, (b) patients' length of stay and overallthroughput, (c) people density over time, and (d) number and location of social interactions.

•• New graphic user interface that provides real-time feedback to the design stakeholders.The new features added in this platform enabled the simulation of real-world hospital designs that include larger and more complex layouts,significantly larger number of occupants (hundreds as opposed to dozens), larger and more context-responsive narratives (26 behaviors

  • grouped in 4 narratives), and longer time span (10 h of simulated time).The new case study supported the pre-occupancy evaluation of comparative design options (not demonstrated in prior work)

 patients and nurses' walking paths and distances, (b) patients' length of stay and overall throughput, (c) people density over time, and (d) number and location of social interactions, previously described in Section 3.1.7. The simulation results are summarized in Table 2.

Figs. 11 and 12 show respectively the walking paths of patients and nurses. In Fig. 11 it can be seen that, in Design B, the patients' flow in front of the doctors' rooms is greater compared to the one in Design A. Fig. 12, instead, reveals that in Design A, the nurses' flow is concentrated in the reception and main waiting areas, while in Design B, it is concentrated in front of the doctors' rooms. The patients' walking distances are shorter in Design B compared to Design A, since the patients can wait in proximity of the nurse and doctor offices. In contrast, nurses' walking distances are shorter in Design A, since one of the nurses' offices is located in front of the doctors' offices.

Aggregated people density and social interactions are displayed respectively in Figs. 13 and 14. In Design A, the zones with the higher density are in proximity to the nurse rooms, while in Design B they are distributed across the clinic, in proximity to both the reception desk and the doctors' rooms. Fig. 14 shows that in Design B, social interactions occur closer to the doctors' offices, compared to Design A, where patients and companions cannot wait in front of the doctors' rooms. In Design A, instead, the social interactions occur in front of the nurses' rooms, potentially interfering with nurses' tasks. Since in Design B the nurses spend more time walking in corridor areas, the chances of social interactions increase compared with Design A, as indicated in Table 2. The maximum and average length of stay is longer in Design B compared to Design A due to a combination of longer nurses walking distances and increased number of social interactions, which delay patients' treatments. Accordingly, the overall patients' throughput time is also longer.

Table 3 summarizes key aspects of the pre-occupancy evaluation of the two design options while accounting for the goals of the stakeholders with respect to each KPI. Specifically, we indicated for each KPIs whether the design stakeholders (i.e. hospital managers, doctors and nurses) aimed to reduce (↓) or increase (↑) the value of the metric. Additionally, for each KPI, we briefly discussed the implications that may emerge while complying or not with the design goals, based on the discussion with the stakeholders and with literature review. Overall, the results reveal that Design A provides better working conditions for doctors since they experience reduced congestion in front of their offices. Additionally, Design A provides better patients' throughput, which improves patients' satisfaction and operational efficiency.

However, it provides less favorable conditions for nurses, since they experience spatial bottlenecks in front of their rooms that could lead to diminished efficiency.

4.3. Discussion

The presented study highlighted the ability of the narrative-based model to support the simulation and evaluation alternative design options in real-world designs of outpatient clinics. Specifically, the model is able to reveal how different spatial configurations impact differently selected performance metrics that depend on human spatial behavior. For instance, the decision on where to sit affects the walking distances to a target, spatial congestion, and the possibility of nurse-patient interactions.

Compared with prior work on narrative-based modeling, our platform exhibits novel features including an extended library of actors, spaces, activities and narratives, an extended narrative manager, and a newly implemented pre-occupancy evaluation module. These features enabled the model scalability and applicability to study human behavior in a significantly more complex setting. Table 4 highlights the newly introduced features.

In addition, the proposed pre-occupancy simulation and evaluation approach exhibits a number of characteristics which distinguishes it from other pre-occupancy multi-agent simulation methods [14,57]. For example, it accounts for (a) collaborative behaviors of multiple actors, (b) the dynamic interactions between planned and unplanned narratives, (c) the dynamic impact that a building design produces on its prospective occupants, and (d) multi-criteria pre-occupancy evaluation metrics visualized in the form of data logs or spatiotemporal data maps. Table 5 summarizes key differences between the proposed approach with other established pre-occupancy evaluation frameworks.

While the proposed study focused on demonstrating the scalability and applicability of narrative-based modeling to simulate and evaluate behavior patterns in real-world outpatient clinics, a number of considerations deserve attention, which pertain to the limitations of the conducted pre-occupancy study. First, we have considered representative narratives in ophthalmology clinics, fixed duration of medical procedures, predefined times to spawn agents, and a fixed number of companions per patients. Second, to generate spaces, actors, activities and narrative libraries we have mostly relied on input from hospital managers, doctors, and nurses. Third, to assess the plausibility of the simulation results we used face validation techniques [68] that involved displaying the simulation results to a panel of experts (i.e. the hospital managers involved in this study). While it cannot be expected to validate simulation results in buildings that have not yet been built, additional studies could be conducted to systematically collect data, calibrate and validate the simulation against human behavior in existing environments. Due to the complex and non-linear nature of human behavior in complex settings, like hospitals, validating human behavior in existing settings may be proven a complicated – albeit necessary – step to improve the simulation accuracy.https://docs.google.com/drawings/u/0/d/sUEfr1Gh38w0zje0JmpQdLQ/image?w=201&h=921&rev=1&ac=1&parent=1wuUtea49E6VO5gbefLg_Ml-nPQNhiIYU

5. Conclusion

In this paper, we tested the scalability and applicability of narrativebased modeling to support the pre-occupancy evaluation of alternative real-world outpatient clinics' designs. Specifically, we developed a simulation and evaluation platform that extends prior work on narrativebased modeling. The platform comprises of the following components: (a) a novel library of spacesactorsactivities and narratives that represent the spaces that people inhabit, the actors that populate the spaces, the activities they perform, and the narratives they are involved in; (b) a refactored narrative manager that coordinates the narratives' unfolding over time; (c) a simulation engine that calculates actors' behavior over; and (d) a newly implemented pre-occupancy evaluation module that calculates an visualizes a list of KPIs measured numerically (e.g., patients' length of stay, walking distances, etc.) or by means of aggregated spatiotemporal data maps (e.g., people density and location of staff-patient social interactions).

We applied the proposed platform to conduct a multi-criteria preoccupancy evaluation study of two alternative designs for an ophthalmology outpatient clinic. The presented case study was conducted in collaboration with two hospitals in Israel. The collaboration involved periodic meetings with the architects, hospital managers and head nurses to collect data, show preliminary results, discuss improvements to the simulation, refine the narrative models, calibrate the model, and define KPI for evaluation with respect to specific design goals. In light of the findings of this study, both architects and hospital managers developed new insights on how the proposed design solution will affect occupancy patterns. For example, they realized that creating an optimal design for doctors may affect nurse procedures, which in turn may impact the overall efficiency.

Beyond extending prior work on narrative-based modeling, the proposed approach exhibits unique features that distinguish it from other established pre-occupancy evaluation methods [14,57]: it accounts for planned and unplanned collaborative behaviors of heterogeneous actors, it reveals the dynamic impact that a building design produces on its prospective occupants, and it calculates multi-criteria pre-occupancy evaluation metrics visualized in the form of data logs or spatiotemporal data maps.

However, the work on narrative-based modeling is still in its initial stage. The model needs further testing in additional case studies beyond hospitals to further demonstrate its scalability and applicability across domains. In addition, further research is needed to demonstrate the effects of narrative-based modeling to support the analysis of collaborative planned and unplanned behaviors in architectural design projects by means of empirical studies. Despite these limitations, the proposed model provides initial steps towards a systematic pre-occupancy evaluation approach to inform architects and other design stakeholders of the impact that a space configuration produces on several context-specific KPI. Specifically, it offers a scalable and modular approach to account for additional narratives at increasing levels of complexity. The library of the proposed narratives, KPI analysis and benchmarks could be further extended to facilitate the simulation of new scenarios.

We believe that further research in this direction could provide architects with a useful method to reduce the uncertainty related to how buildings will be used by future occupants and will possibly lead to

designs that better support human needs and perform better operationally.

Acknowledgments

This research was supported by the European Research Council grant (FP-7 ADG 340753), the Israel Science Foundation grant 323/15, the Azrieli Foundation, and the Jacobs scholarship. The case study

Appendix A

Table A1

Element

Semantic

Property type

Property name

Description

Zone

All

Static

ID

Unique identification of the zone

  

Dynamic

List of agents

Agents located within the zone at a given time

   

People density

Agents' number divided by zone sqm

 

Waiting

Static

List of sitting spots

All predefined sitting spots

   

List of standing spots

All predefined standing spots

   

List of ED spots

All predefined sitting spots for ED patients

  

Dynamic

Available sitting spots

Currently available sitting spots

   

Available standing spots

Currently available standing spots

   

Available ED spots

Currently available ED spots

 

Queuing

Static

List of queuing spots

All predefined queuing spots

  

Dynamic

Available spots

Currently available queuing spots

 

Clinic

Static

Patient spot

Location of where the patient should stay in the room

   

Companion spot

Location of where the companion should stay

   

Staff spot

Location of where the staff member should stay

   

Staff associated

ID of the staff member allocated to the clinic room

  

Dynamic

Occupancy status

Presence (or absence) of the staff member in the room

   

List of waiting patients

Patients waiting to be treated by the staff member

   

Ready for patient

“True” if the staff is ready to receive the patient

Spot

All

Static

ID

Unique identification of the spot

   

Current zone

Zone in which the spot is located

  

Dynamic

Occupied

Whether it is busy or occupied by an actor

   

Actor

ID of the actor that occupies the spot

 

Waiting

Static

Sit/stand

Indicates if people can sit or should stand in this spot

 

Queuing

Static

Next queuing spot

Stores the location of the next queuing spot

Table A2

Actor profiles.

    

Role

Property type

Property name

Description

  

All

Static

ID

Unique identification of the actor

 
  

Movement speed

Default agent speed [1.4 m/s]

 
 

Dynamic

Current zone

Current zone an actor is located in

 
  

Current activity

Current activity the agent is involved in

 
  

Movement target

Current target the actor is moving towards

 

Patient

Static

Patient type

Regular or ED

 
  

Doctor ID

ID of the doctor associated with the patient

 
  

Nurse ID

ID of the nurse associated with the patient

 
  

Companion ID

ID of the companion associated with the patient

 
 

Dynamic

Waiting spot

Spot where the actor is currently waiting

 
  

Queue for doctor

Position in the queuing list for the doctor

 
  

Queue for nurse

Position in the queuing list for the nurse

  

Available for interactions

“True” if actor can be involved in staff-patient interactions. This depends on the current activity of the actor and whether it was recently involved in a social interaction

Companion

Static

Patient ID

ID of the patient it is associated with

 

Dynamic

Waiting spot

Spot where the actor is currently waiting

Nurse

Static

Office room

ID of the nurse's office room

 

Dynamic

Latest # of patient visited

Number of patients visited since the nurse left their room to interact with a doctor. This value is used as precondition for the narrative described Table A6.

  

Available for interactions

“True” if actor can be involved in staff-patient interactions. The status of this property depends on the current activity of the actor and whether it was recently involved in a social interaction

Doctor

Static

Office room

ID of the doctor's office room

Secretary

Static

Work station

ID of the workstation

Space profiles.

presented in this paper has been conducted at the Technion – Israel Institute of the Technology. We are grateful to Clalit HMO management, Meir Medical Center management and staff, Rambam Health Care Campus management and staff, architect Faten Tabry Kattouf, and Mochly-Eldar Architects for their collaboration. We thank Dr. Efrat Eizenberg, Dr. Lusi Morhaim, Dr. Mor Shilon and Ms. Dalia Arussy for collecting data in the hospitals.

Table A3

Activity types.

Name

Parameter list

Description

Move

Actor list

List of actors participating in the activity

 

Target

Movement target

Interact

Actor list

List of actors participating in the activity

 

Duration

Activity duration

Wait

Actor list

List of actors participating in the activity

 

Waiting spot

Spot where the actor waits

 

End condition

Either a change in the state of the world or a specific duration

Queue

Actor list

List of actors participating in the activity

 

Queuing zone

Zone where the agent is queuing

Table A4

Regular patient flow narrative algorithm.

https://docs.google.com/drawings/u/0/d/sxoxmGlC6eToSIHlUmrXloA/image?w=693&h=1&rev=1&ac=1&parent=1wuUtea49E6VO5gbefLg_Ml-nPQNhiIYU

Input: Regular Patient (P), Companion (C), Nurse (N), Doctor (D)

Preconditions: Regular Patient and Companion enter the ward Performing steps:

Step Activity

Parameters

Description

1

Move

P, C

Reception Queuing

Zone

The patient and companion go to a queuing zone in front of the reception desk.

2

Queue

P, C

The patient and companion queue for the secretary with the shortest queue. This information is calculated by the Queuing Zone and

  

Reception Queuing Zone

communicated to the narrative.

3

Interact

P, C, Secretary (S)

2.5 min

The patient and companion discuss with the secretary for a fixed duration

4

Move

P, C

The narrative first calculates the most favorable waiting spots close to the nurse area for both actors, then it initiates the Move function

  

Waiting Spots

with that specific location. As result, the patient and visitor go to wait to their designated spots.

5

Wait

P, C

The patient and companion wait for their turn to be received by the nurse. The queuing list is stored within the Nurse office zone and it is

  

Waiting Spots

Nurse Treatment List

queried by the waiting activity.

6

Move

P, C

Nurse Treatment

Room

The patient and companion go to the nurse treatment room.

7

Interact

Patient, Companion,

Nurse

7 min

The nurse performs the visual acuity test to the patient

8

Move

P, C

Waiting Spots

The narrative calculates new waiting spots for the group. Then, patient and companion go to wait for the eye dilation in those spots.

9

Wait

P, C

Waiting Spots

30 mins

The patient and companion wait for a specific amount of time in their waiting spots until the eyes have been dilated.

10

Wait

P, C

The patient and companion wait for their turn to be called by a doctor. The queuing list is stored within the Doctor office zone and it is

  

Doctor Treatment

List

queried by the waiting activity.

11

Move

P, C

Doctor Room

The patient and companion go to the doctor room.

12

Interact

P, C, D 20 min

The doctor treats the companion for a specific amount of time.

13

Move

Patient, Companion Secretary Zone

The patient and companion are directed towards a queuing zone in front of the secretary's desk.

14

Queue

P, C

Secretary Zone

The patient and companion queue in front of the secretary's reception.

15

Interact

P, C, S 2.5 min

The patient and companion discuss with the secretary for a given duration to complete the checkout

16

Move

P, C

The patient and companion go to a randomly selected exit door

Exit Zone

Postconditions: Patient and Companion leave the ward

https://docs.google.com/drawings/u/0/d/sTZjt-VpIs9gL7-Z72w5dUQ/image?w=693&h=1&rev=1&ac=1&parent=1wuUtea49E6VO5gbefLg_Ml-nPQNhiIYU

Table A5

ER patient flow narrative algorithm.

https://docs.google.com/drawings/u/0/d/s3VLWE7FJ8Pqkci4TbfmEVg/image?w=693&h=1&rev=1&ac=1&parent=1wuUtea49E6VO5gbefLg_Ml-nPQNhiIYU

Input: ED Patient (P), Companion (C), Doctor (D) Preconditions: ED Patient and Companion enter the ward Performing steps:

Step

Activity

Parameters

Description

1

Move

P, C

ED Waiting Spots

The patient and companion go to the available ED waiting spots calculated by the narrative

2

Wait

P, C

Doctor Treatment List

The patient and companion wait for their turn to be received by a doctor.

(continued on next page)

Table A5 (continued)

3 Move

P, C

Doctor Room

The patient and visitor go to the doctor treatment room.

4 Interact

P, C, D 20 mins

The doctor treats the patient.

5 Move

P, C

The patient and visitor go to a random exit door

Exit Zone Postconditions: ED Patient and Companion leave the ward

 

Table A6

Nurse-doctor interaction narrative algorithm.

 

Input: Nurse (N), Number of Patients visited before administrative work (PV)

Preconditions: Latest Number of Patients Visited (LV) == PV Performing steps:

Step Activity Parameters

Description

1 Move

N

Doctor Office

The nurse goes to a doctor room randomly selected by the narrative.

2 Interact

N, D 2 mins

The nurse interacts with the doctor.

3 Move

P, C

The nurse goes back to their room

Postconditions: LV = 0

Exit Zone

 

Table A7

Staff-patient interaction narrative algorithm.

  

Input: Patient (P), Nurse (N)

Preconditions: Patient and Nurse are in proximity of each other + the nurse have not engaged in a social interaction for the previous 3 mins, and the patient has not engaged in a social interaction for the previous 30 mins Performing steps:

Step Activity Parameters Description

  1. Move P, N The narrative directs a patient and a nurse to move towards each other.

Middle point between actors

  1. Interact C, N The nurse and the companion interact for a stipulated duration of 30 s

30 s

Postconditions: Nurse cannot engage in a social interaction for the next 3 mins, and the patient cannot engage in social interaction for the next 30 mins

https://docs.google.com/drawings/u/0/d/sxJ-_9KWPgH2YncnCpmKsuQ/image?w=693&h=1&rev=1&ac=1&parent=1wuUtea49E6VO5gbefLg_Ml-nPQNhiIYU

References

  1. D. Yan, W. O'Brien, T. Hong, X. Feng, H. Burak Gunay, F. Tahmasebi, A. Mahdavi, Occupant behavior modeling for building performance simulation: current state and future challenges, Energy and Buildings 107 (2015) 264–278 https://doi.org/10.

1016/j.enbuild.2015.08.032.

  1. J. Zeisel, Inquiry by design: tools for environment-behaviour research. CUP archive, Retrieved March 24, 2019 from, 1984. https://www.google.com/books?hl=en& lr=&id=AwU4AAAAIAAJ&oi=fnd&pg=PA1&dq=zeisel+inquiry+by+design +1984&ots=o5xUh9VhST&sig=cLF6SU_0aVywHBsIIEF-Gs5YAsY.
  2. B. Lawson, What Designers Know, Architectural Press, 2004, https://doi.org/10.

4324/9780080481722.

  1. H.W. Rittel, On the Planning Crisis: Systems Analysis of the “First and Second Generations”, Institute of Urban and Regional Development Berkeley, 1972 Retrieved March 24, 2019 from http://courses.cs.vt.edu/cs4634/reading/Rittel. pdf.
  2. A. Andriamamonjy, D. Saelens, R. Klein, An automated IFC-based workflow for building energy performance simulation with Modelica, Autom. Constr. 91 (2018) 166–181 https://doi.org/10.1016/j.autcon.2018.03.019.
  3. L. Xinzheng, Y. Ning, J. Jianjing, Application of computer simulation technology for structure analysis in disaster, Autom. Constr. 13 (5) (2004) 597–606 https://doi. org/10.1016/j.autcon.2004.04.002.
  4. L. Ben-Alon, R. Sacks, Simulating the behavior of trade crews in construction using agents and building information modeling, Autom. Constr. 74 (2017) 12–27 https://doi.org/10.1016/j.autcon.2016.11.002.
  5. L. Bellia, F. Fragliasso, Evaluating performance of daylight-linked building controls during preliminary design, Autom. Constr. 93 (2018) 293–314 https://doi.org/10. 1016/j.autcon.2018.05.026.
  6. W. Yan, Y.E. Kalay, Simulating the behavior of users in built environments, Journal of Architectural and Planning Research (2004) 371–384 https://www.jstor.org/ stable/43030713.
  7. X. Zheng, T. Zhong, M. Liu, Modeling crowd evacuation of a building based on seven methodological approaches, Build. Environ. 44 (3) (2009) 437–445 https:// doi.org/10.1016/j.buildenv.2008.04.002.
  8. Kapadia, M., Pelechano, N., Allbeck, J., & Badler, N. (2015). Virtual crowds: steps toward behavioral realism. Synthesis Lectures on Visual Computing: Computer Graphics, Animation, Computational Photography, and Imaging, 7(4), 1–270. https://doi.org/10.2200/S00673ED1V01Y201509CGR020.
  9. P. Boguslawski, L. Mahdjoubi, V. Zverovich, F. Fadli, A dynamic approach for evacuees' distribution and optimal routing in hazardous environments, Autom.

Constr. 94 (2018) 11–21 https://doi.org/10.1016/j.autcon.2018.05.032.

  1. S.W. Hong, Y.G. Lee, The effects of human behavior simulation on architecture major students' fire egress planning, Journal of Asian Architecture and Building Engineering 17 (1) (2018) 125–132 https://doi.org/10.3130/jaabe.17.125.
  2. W. Shen, Q. Shen, Q. Sun, Building information modeling-based user activity simulation and evaluation method for improving designer–user communications, Autom. Constr. 21 (2012) 148–160 https://doi.org/10.1016/j.autcon.2011.05.022.
  3. V. Tabak, B. de Vries, J. Dijkstra, Simulation and validation of human movement in building spaces, Environment and Planning B: Planning and Design 37 (4) (2010) 592–609 https://doi.org/10.1068/b35127.
  4. D. Schaumann, S. Breslav, R. Goldstein, A. Khan, Y.E. Kalay, Simulating use scenarios in hospitals using multi-agent narratives, J. Build. Perform. Simul. 10 (5–6) (2017) 636–652 https://doi.org/10.1080/19401493.2017.1332687.
  5. Kalay, Y. E. (2004). Architecture's New Media: Principles, Theories, and Methods of Computer-aided Design. Cambridge, MA: MIT Press. ISBN-10: (0262112841).
  6. H.A. Simon, The Sciences of the Artificial, MIT Press, Cambridge, MA, 1969 (ISBN-

10: 9780262691918).

  1. H.W. Rittel, M.M. Webber, Dilemmas in a general theory of planning, Policy. Sci. 4 (2) (1973) 155–169 Retrieved March 24, 2019 from https://link.springer.com/ article/10.1007/BF01405730.
  2. H.-B. Seo, Y.-S. Choi, C. Zimring, Impact of hospital unit design for patient-centered care on nurses' behavior, Environ. Behav. 43 (4) (2011) 443–468 https://doi.org/ 10.1177/0013916510362635.
  3. E.K. Peavey, J. Zoss, N. Watkins, Simulation and mock-up research methods to enhance design decision making, HERD: Health Environments Research & Design Journal 5 (3) (2012) 133–144 https://doi.org/10.1177/193758671200500313.
  4. S. Kumar, M. Hedrick, C. Wiacek, J.I. Messner, Developing an experienced-based design review application for healthcare facilities using a 3D game engine, Journal of Information Technology in Construction 16 (2011), http://www.itcon.org/ 2011/6.
  5. M. Kim, X. Wang, P. Love, H. Li, S.-C. Kang, Virtual reality for the built environment: a critical review of recent advances, Journal of Information Technology in Construction 18 (2013) (2013) 279–305 http://www.itcon.org/ 2013/14.
  6. Hillier, B., & Hanson, J. (1984). The Social Logic of Space. Cambridge: Cambridge University Press. ISBN-10: (0521367840).
  7. C. Hölscher, S. Büchner, T. Meilinger, G. Strube, Map use and wayfinding strategies in a multi-building ensemble, Spatial Cognition V Reasoning, Action, Interaction, 2007, pp. 365–380, , https://doi.org/10.1007/978-3-540-75666-8_21.
  8. M. Bhatt, A. Borrmann, R. Amor, J. Beetz, Architecture, computing, and design assistance, Autom. Constr. 32 (2013), https://doi.org/10.1016/j.autcon.2013.01. 001.
  9. M. Bhatt, C. Schultz, M. Huang, The shape of empty space: human-centred cognitive foundations in computing for spatial design, IEEE Symposium on Visual Languages and Human-centric Computing, 2012, pp. 33–40 https://doi.org/10.1109/VLHCC. 2012.6344477.
  10. K. Sailer, A. Budgen, N. Lonsdale, A. Turner, A. Penn, Pre and post occupancy evaluations in workplace environments, Journal of Space Syntax 1 (1) (2010) 199–213 Retrieved March 24, 2019 from http://joss.bartlett.ucl.ac.uk/journal/ index.php/joss/article/view/27.
  11. R.-J. Dzeng, W.-C. Wang, F.-Y. Hsiao, Y.-Q. Xie, An activity-based simulation model for assessing function space assignment for buildings: a service performance perspective, Computer-Aided Civil and Infrastructure Engineering 30 (12) (2015) 935–950 https://doi.org/10.1111/mice.12177.
  12. Eastman, C. M., & Siabiris, A. (1995). A generic building product model incorporating building type information. Autom. Constr., 3(4), 283–304. https://doi. org/10.1016/0926-5805(94)00028-L.
  13. A. Ekholm, Modelling of user activities in building design, Proceedings of 19th Education and Research in Computer Aided Architectural Design in Europe Conference, 2001, pp. 67–72 Helsinki. Retrieved March 24, 2019, from http:// papers.cumincad.org/cgi-bin/works/Show?c596.
  14. T.W. Kim, M. Fischer, Automated generation of user activity–space pairs in spaceuse analysis, J. Constr. Eng. Manag. 140 (5) (2014) 04014007https://doi.org/10.

1061/(ASCE)CO.1943-7862.0000839.

  1. T.W. Kim, M. Fischer, Ontology for representing building users' activities in spaceuse analysis, J. Constr. Eng. Manag. 140 (8) (2014) 04014035https://doi.org/10.

1061/(ASCE)CO.1943-7862.0000881.

  1. Maher, M. L., Simoff, S. J., & Mitchell, J. (1997). Formalising building requirements using an activity/space model. Autom. Constr., 6(2), 77–95. https://doi.org/10.

1016/S0926-5805(96)00171-9.

  1. G. Wurzer, Schematic systems—constraining functions through processes (and vice versa), Int. J. Archit. Comput. 8 (2) (2010) 201–217 https://doi.org/10.1260/14780771.8.2.201.
  2. A. Çekmiş, I. Hacıhasanoğlu, M.J. Ostwald, A computational model for accommodating spatial uncertainty: predicting inhabitation patterns in open-planned spaces, Build. Environ. 73 (2014) 115–126 https://doi.org/10.1016/j.buildenv. 2013.11.023.
  3. Y. Cao, T. Wang, X. Song, An energy-aware, agent-based maintenance-scheduling framework to improve occupant satisfaction, Autom. Constr. 60 (2015) 49–57 https://doi.org/10.1016/j.autcon.2015.09.002.
  4. J.B. Kim, W. Jeong, M.J. Clayton, J.S. Haberl, W. Yan, Developing a physical BIM library for building thermal energy simulation, Autom. Constr. 50 (2015) 16–28 https://doi.org/10.1016/j.autcon.2014.10.011.
  5. D. Helbing, P. Molnar, Social force model for pedestrian dynamics, Phys. Rev. E 51 (5) (1995) 4282 https://doi.org/10.1103/PhysRevE.51.4282.
  6. L.F. Henderson, The statistics of crowd fluids, Nature 229 (1971) 381–383 https:// doi.org/10.1038/229381a0.
  7. Hoogendoorn, S. P., & Bovy, P. H. (2004). Pedestrian route-choice and activity scheduling theory and models. Transp. Res. B Methodol., 38(2), 169–190. https:// doi.org/10.1016/S0191-2615(03)00007-9.
  8. R.L. Hughes, The flow of human crowds, Annu. Rev. Fluid Mech. 35 (1) (2003) 169–182 https://doi.org/10.1146/annurev.fluid.35.101101.161136.
  9. A. Treuille, S. Cooper, Z. Popović, Continuum crowds. In ACM Transactions on Graphics (TOG) (Vol. 25, pp. 1160–1168). ACM, Retrieved March 24, 2019 from, 2006. http://dl.acm.org/citation.cfm?id=1142008.
  10. S. Bandini, M.L. Federici, G. Vizzari, Situated cellular agents approach to crowd modeling and simulation, Cybernetics and Systems: An International Journal 38 (7) (2007) 729–753, https://doi.org/10.1080/01969720701534141.
  11. V. Blue, J. Adler, Cellular automata microsimulation of bidirectional pedestrian flows, Transportation Research Record: Journal of the Transportation Research Board 1678 (1999) 135–141 https://doi.org/10.3141/1678-17.
  12. Y.N. Marmor, B. Golany, S. Israelit, A. Mandelbaum, Designing patient flow in emergency departments, IIE Transactions on Healthcare Systems Engineering 2 (4) (2012) 233–247 https://doi.org/10.1080/19488300.2012.736118.
  13. Løvås, G. G. (1994). Modeling and simulation of pedestrian traffic flow. Transp. Res. B Methodol., 28(6), 429–443. https://doi.org/10.1016/0191-2615(94) 90013-2.
  14. E. Bonabeau, Agent-based modeling: methods and techniques for simulating human systems, Proc. Natl. Acad. Sci. 99 (3) (2002) 7280–7287, https://doi.org/10.1073/ pnas.082080899.
  15. M. Wooldridge, N.R. Jennings, Intelligent agents: theory and practice, Knowl. Eng.

Rev. 10 (02) (1995) 115–152 https://doi.org/10.1017/S0269888900008122.

  1. D. Helbing, I.J. Farkas, P. Molnar, T. Vicsek, Simulation of pedestrian crowds in normal and evacuation situations, Pedestrian and Evacuation Dynamics 21 (2) (2002) 21–58 Retrieved March 24, 2019, from https://pdfs.semanticscholar.org/ 6b72/b31c2531b2a910ecbe6baa095b0907a7448a.pdf.
  2. M.L. Chu, P. Parigi, K. Law, J.-C. Latombe, Modeling social behaviors in an evacuation simulator, Computer Animation and Virtual Worlds 25 (3–4) (2014) 373–382 https://doi.org/10.1002/cav.1595.
  3. X. Pan, C.S. Han, K. Dauber, K.H. Law, A multi-agent based framework for the simulation of human and social behaviors during emergency evacuations, AI & Soc. 22 (2) (2007) 113–132 Retrieved March 24, 2019 from https://link.springer.com/ article/10.1007/s00146-007-0126-1.
  4. F. Tang, A. Ren, GIS-based 3D evacuation simulation for indoor fire, Build. Environ.

49 (2012), https://doi.org/10.1016/j.buildenv.2011.09.021.

  1. D. Zhao, J. Wang, X. Zhang, X. Wang, A cellular automata occupant evacuation model considering gathering behavior, International Journal of Modern Physics C 26 (08) (2014) 1550089 https://doi.org/10.1142/S0129183115500898.
  2. Cassol, V. J., Testa, E. S., Jung, C. R., Usman, M., Faloutsos, P., Berseth, G., …

Musse, S. R. (2017). Evaluating and optimizing evacuation plans for crowd egress. IEEE Comput. Graph. Appl., 37(4), 60–71. doi:https://doi.org/10.1109/MCG.2017. 3271454.

  1. D. Thalmann, S.R. Musse, Crowd Simulation, Springer London, London, 2013 Retrieved March 24, 2019 from http://link.springer.com/10.1007/978-1-44714450-2.
  2. S. Shin, S. Jeong, J. Lee, S.W. Hong, S. Jung, Pre-occupancy evaluation based on user behavior prediction in 3D virtual simulation, Autom. Constr. 74 (2017) 55–65 https://doi.org/10.1016/j.autcon.2016.11.005.
  3. W. Shen, X. Zhang, G. Qiping Shen, T. Fernando, The user pre-occupancy evaluation method in designer–client communication in early design stage: a case study, Autom. Constr. 32 (Supplement C) (2013) 112–124, https://doi.org/10.1016/j. autcon.2013.01.014.
  4. D. Simeone, Y.E. Kalay, D. Schaumann, Using game-like narratives to simulate human behavior in built environments, Proceedings of the 18th International Conference on Computer-Aided Architectural Design Research in Asia, 2013, pp.

199–208 Singapore. Retrieved March 24, 2019 from https://www.researchgate. net/profile/Davide_Simeone/publication/275330352_USING_GAME-LIKE_ NARRATIVE_TO_SIMULATE_HUMAN_BEHAVIOUR_IN_BUILT_ENVIRONMENTS/ links/553774b50cf2058efdeabe45/USING-GAME-LIKE-NARRATIVE-TOSIMULATE-HUMAN-BEHAVIOUR-IN-BUILT-ENVIRONMENTS.pdf.

  1. A. Shoulson, D. Garcia, N.I. Badler, Selecting Agents for Narrative Roles. AAAI Technical Report WS-11-18, Retrieved March 24, 2019 from, 2011. https://www. aaai.org/ocs/index.php/AIIDE/AIIDE11WS/paper/viewPaper/4089.
  2. F. Becker, S. Douglass, The ecology of the patient visit: physical attractiveness, waiting times, and perceived quality of care, The Journal of Ambulatory Care Management 31 (2) (2008) 128–141 https://doi.org/10.1097/01.JAC.

0000314703.34795.44.

  1. R. Ulrich, X. Quan, C. Zimring, A. Joseph, R. Choudhary, The role of the physical environment in the hospital of the 21st century: a once-in-a-lifetime opportunity. The Center for Health Design, Retrieved March 24, 2019 from, 2004. http://www. designandinstitution.com/s/Role-Physical-Environ-in-the-21st-Century-Hospital_0. pdf.
  2. C.M. Zimring, R.S. Ulrich, X. Zhu, J.R. DuBose, H.-B. Seo, Y.-S. Choi, ... A. Joseph, A review of the research literature on evidence-based healthcare design, HERD: Health Environments Research & Design Journal 1 (3) (2008) Retrieved March 24, 2019 from https://smartech.gatech.edu/handle/1853/25676.
  3. M. Brodeschi, N. Putievsky Pilosof, Y.E. Kalay, The definition of semantic of spaces in virtual built environments oriented to BIM implementation, Computer Aided Architectural Design Futures. Proceedings, 2015, pp. 331–346 Retrieved March 24, 2019 from http://cumincad.scix.net/data/works/att/cf2015_331.content.pdf.
  4. Sopher, H., Schaumann, D., & Kalay, Y. E. (2016). Simulating human behavior in (un) built environments: using an actor profiling method. World Academy of Science, Engineering and Technology, International Journal of Computer,

Electrical, Automation, Control and Information Engineering, 10(12), 2074–2083.

Retrieved March 24, 2019 from https://www.semanticscholar.org/paper/ Simulating-Human-Behavior-in-(-Un-)-Built-%3A-Using-Sopher-Schaumann/ 0d9b2cbfe487dc811bee280f579083ccd868f2c0.

  1. J.I. Westbrook, A. Woods, M.I. Rob, W.T. Dunsmuir, R.O. Day, Association of interruptions with an increased risk and severity of medication administration errors, Arch. Intern. Med. 170 (8) (2010) 683–690, https://doi.org/10.1001/ archinternmed.2010.65.
  2. M. Morad, E. Zinger, D. Schaumann, N. Putievsky Pilosof, Y. Kalay, A dashboard model to support spatio-temporal analysis of simulated human behavior in future built environments, Symposium on Simulation for Architecture and Urban Design, 2018 Retrieved March 24, 2019 from https://www.researchgate.net/publication/ 325871248_A_Dashboard_Model_to_Support_Spatio-Temporal_Analysis_of_ Simulated_Human_Behavior_in_Future_Built_Environments.
  3. R.G. Sargent, Verification and validation of simulation models, Journal of Simulation 7 (1) (2013) 12–24 https://doi.org/10.1057/jos.2012.20

    Received 12 December 2018; Received in revised form 29 March 2019; Accepted 27 June 2019

    0926-5805/ © 2019 Elsevier B.V. All rights reserved.

Tags: Reference
     
Copywrite © 2020 LOGICMOO (Unless otherwise credited in page)