NIVEL 3 DE DESEMPEÑO PISA CIENCIAS
XI.3. Primer periodo escolar, al concluir el tercer grado de preescolar, entre 5 y 6 años de edad 1 Estándares de Español
XI.3.2. Aprendizajes esperados de Lenguaje y comunicación
An important component in the occupancy estimation framework proposed in this
dissertation is the model which serves as prior knowledge of the patterns of occupants’ movement so that location of occupants can be inferred when data from sensors are not reliable.
There are majorly two categories of models that can serve this purpose. An agent based model
models each individual occupant as an autonomous agent with its own rules of performing
behaviors and interacting with others. A coarse model models the occupants as entities flowing
in a network of nodes, where each node represents a room or a corridor and each edge represents
the connectivity among the rooms and corridors. In this section, we give a literature review on
2.2.1 Agent-based models
Agent-based model is a useful and prominent tool to study people’s movement behavior in
various environments. Numerous research on building such model have been conducted by
previous researchers. In [49], the author builds an agent-based model that simulates the
movement behaviors of pedestrian in an urban environment with facilities such as church, park,
monument, post office and library. The agent-based model is featured with two different
behaviors. One is random walk, meaning that the agent moves on the street casually without a
clear purpose; the other is purposely movement, meaning that the agent has a predefined
destination to move towards to. Experiments show that in the random walk case there is strong
correlation between local integration and movements of pedestrian where this correlation is weak
in the purposely walk case. In [50], the author builds an agent-based model based on
psychological factors. Specifically, the psychological factors are classified into two classes,
psychophysical factor and psychosocial factor. The psychophysical factors include the range of the agent’s vision and the variation of their moving speed. The psychosocial factor include the goal for agent’s movement, for example, to reach a predefined destination or to avoid other agents; and the preference the agent chooses to accomplish his goal as fast as possible. The
behaviors of an agent are further categorized into three phases, one is strategic phase, another is
tactical phase and the final one is the operational phase. In strategic phase, a global destination is
determined for each of the agent so that agents always try to move towards their destinations. In
tactical phase, the agents try to avoid other agents and obstacles based on the psychological
factors introduced before. In the operational phase, the agent calculates the speed and direction
based on decision made by previous two phases to determine its movement. This model can
for simulating crowd at Tawaf area is presented. This model is featured with two levels of
movement of pedestrian, the macroscopic movements and microscopic movements. The
macroscopic movements define the way finding behaviors or navigation behaviors of pedestrians
while the microscopic movements define the interactions among pedestrians when they are
performing macroscopic behaviors. Such interactions include avoiding each other or help each
other, etc. The microscopic movements are carried out using a cellular automata and the
macroscopic movements are implemented using path tables. This work is still at its early stage
and author plans to implement several simulator in the later stages of the proposed work. In [52],
author proposes a layered approach to model the dynamics of pedestrian crowd. The surface of
the 2D environment is divided into different layers to indicate the occupancy, the position of
static obstacles and possibly the dynamic obstacles situated in the environment. These layers
contain the necessary information for agent to consult for navigation purposes. Finally, the agent
utilizes the layered environment and uses Markov decision process and semi-Markov decision
process approach to find the correct move given its occupying cell. In [53], author builds an
agent based model based on cellular automata to simulate pedestrian’s behavior. This approach uses a matrix to represent the agents’ preferred movements on all directions and uses dynamic floor field to model the trace of the pedestrians’ movement. It uses static field to represent attractive point such as exit on the map. To model the comfortable distance the pedestrian seeks
to keep from others, the author uses the concept of repulsion field which defines the force the
pedestrian agent puts upon its neighbor cells. These concepts together determines the transition
probability an agent moves from one cell to another. In [54], author proposes a model for
simulating the pedestrian crowd dynamics. This model is featured with an environment model
top layer, a route map is utilized to represent the structure and connectivity among regions in the
map. The second layer consists of a navigation map describing the paths to round over obstacles
in each region. At the lowest layer, the description about the information of each objects is
presented. The route map is a topological graph implemented with entries and exits in the regions
representing the nodes of the graph. The distance between the nodes of the graph becomes the weight of the edge of the graph. To decide a path of an agent, the agent’s destination and position are added to the topological graph and a shortest path algorithm is utilized over the graph to find
a suitable path for the agent. This approach is similar to the approach to plan path and navigation
for the agent proposed in this dissertation. Author in [54] also utilizes social force theory to
model the avoidance and interaction between agents.
In most of the work listed in this section, the agent-based model is constructed with two
different components, one is a model for the environment including obstacles and topological
structure of the environment, the other is a model for the decision and action of the agent
including macroscopic behavior such as navigation and microscopic behavior such as avoiding
each other. Inspired by these works, in this dissertation, the agent-based model proposed as a
high-resolution and low abstraction level model also contains a model for describing the
properties of the environment and a model to define the agent’s behavior such as navigation and avoiding each other. Furthermore, we adopt a rule-based approach to define the agents’ decision
on how to perform the behaviors under different circumstances.
2.2.2 Coarse Models
Coarse models in pedestrian simulation include those model the pedestrian moving
dynamics as a queuing network and those model the environment as a graph. It yields great
Some previous research have been conducted in building and utilizing coarse models. In [55],
the author discusses both of the drawback of using merely coarse grained network models and
merely fine grained network models to simulate the occupants’ behaviors in evacuation scenario and proposes to combine the two approaches together to increase the accuracy of the model and
at the same time reduce computational cost. In the simulation, the unprotected area are modeled
using fine network. In the fine network modeling approach, the whole area is divided into a
number of grid cells. Each cell can be occupied by exactly one occupant. By doing this, the
author aims at simulating the interactions among occupants and the impact of the interactions on
evacuation speed of the evacuees. The protected area is modeled using coarse network, where
each node of the network represents a room or a segment of the corridor. Since movements of
evacuees in the protected network are more orderly organized, simulation of interactions among
evacuees are not necessarily needed. Instead of using a coordinate to indicate the occupant’s
position as in fine node, the coarse node uses a one dimension value named “distance” to represent the position of the occupant. Experiment shows that the integrated approach has better
computational performance than using just fine network and can represent more details than
using just coarse network approach. In [56], the author develops a model based on network to
reproduce the pedestrian’s behaviors such as collection, spillback and dissipation. The target environment is divided into grid cells with hexagon shapes. A variable called “potential” is
utilized to indicate the distance of each cell to the exit and the minimum number of movements
an occupant must perform to reach the exit. The grid cell representation of the building structure
is then translated into a graph whose vertex represents each cell in the hexagon grid and edge
represents the pedestrian flow from one cell to another. The directed edges of the graph are
potential to a vertex of cell that has a lower potential. The result of simulations reveals that this
model can realistically simulate the behavior of collection, spillback and dissipation of
pedestrians. In[57], author proposes to use state dependent queuing network to represent and
simulate the flow dynamics of the pedestrians. The author proposes to divide modeling
procedure into three different stages: the representation stage, analysis stage and synthesis stage.
In the representation stage, the network topology of the underlying building or facility is created
with the rooms and corridors being treated as the nodes in the network. In the analysis stage, the
performance characteristics of the queuing network is determined. In the final stage of synthesis,
once the performance characteristics are determined in the analysis stage, the performance of a
queuing network can be assessed and evaluated to solve various design problem.
Coarse models listed in this section normally involves modeling the environment as a
graph with edges and vertex or a network with routes and nodes. The benefit of doing this is to
reduce the computational complexity of the simulation. Coarse model is suitable in the condition
that a large number of pedestrian or occupants are involved. Under such circumstances, agent-
based model may require more computational resources which sometimes is impossible. In this
dissertation, we proposed a simple graph model which models the building structure as a set of
vertex and edges, and models the occupants as flows in the network. This model is more efficient
than the agent-based model when large number of occupants get involved.