18-03-2014, 12:35 PM
Enhancing Parking Simulations Using Peer-Designed Agents
[attachment=60176]
Abstract
In this paper, we investigate the usefulness of peer-designed
agents (PDAs) as a turn-key technology for enhancing parking simulations.
The use of PDAs improves the system’s ability to capture the dynamics
of the interaction between individuals in the system, each theoretically
exhibiting a different strategic behavior. Furthermore, since people in
general are inherently rational and computation bounded, simulating this
domain becomes even more challenging. The advantage of PDAs in this
context lies in their ability to reliably simulate a large pool of human
individuals with diverse strategies and goals. We demonstrate the efficacy
of the proposed method by developing a large-scale simulation system
for the parking space search domain, which plays an important role in
urban transport systems. The system is based on 34 different parking
search strategies. Most of these strategies are substantially different from
synthetic strategies that are used in prior literature. A quantitative analysis
of the PDAs indicates that they reliably capture their designers’ real-life
strategies. Finally, we demonstrate the usefulness of PDA-based parking
space search simulation by utilizing it to evaluate four different informa-
tion technologies that are of increasing use in recent years.
INTRODUCTION
Simulation is an important tool for studying emergent behavior.
Using simulation, a researcher can carry out extensive data collection
in a simple, safe, and inexpensive way. While, in real transport
systems, the cost of a system evaluation is substantial, monetary, or
otherwise, the use of simulation enables evaluating various variants
of the system quickly and simply “with the click of a button.” Not
surprisingly, in recent years, agent technology in artificial intelligence
has become the dominating approach to developing simulation systems
and large-scale distributed systems [1]. While agents’ capabilities
substantially improve the modeling of the individuals that they repre-
sent, researchers have noted the challenge of constructing simulation
systems in domains where key players are people [2], particularly due
to their diverse behavior patterns.
RELATED WORK
Parking space search is an important problem for urban planners and
has been addressed by many traffic and transportation researchers [8],
[9]. Previous work in the area of parking simulations has placed little
emphasis on extracting user parking strategies, and most simulations
reported in the literature use a limited set of preprogrammed strategies.
For example, in PARKSIM, a network model is used to represent
the driver’s choice of behavior and consequently compare parking
facility layouts [10]. Other studies evaluate alternative parking system
designs and operating policies [11]. Recently, Jonkers et al. [9] have
recognized that people follow many different strategies when looking
for a parking space, yet their simulation focused on just two strategies.
To the best of our knowledge, no previous research in the area of
simulated parking lots incorporates a large collection of realistic
parking strategies as part of the entity design, as proposed in this paper.
P EER -D ESIGNED AGENT-BASED S IMULATION
The main idea in PDA-based design is to separate the logic of the
simulated agent’s strategy from the generic functionalities required to
have the agent participate in the simulation. This way, the strategy
development task can be distributed and delegated to a population that
is as similar as possible to the simulated population. We refer to these
people as strategy designers as they embed their strategies in PDAs.
The simulation is controlled and dynamic, and allows the use of
various agents of different types. Different simulation parameters can
be varied as well (e.g., using different parking lot structures, lanes,
and driving directions). The simulation also stores, at any given time,
the system’s current state (e.g., the location of cars and the location of
occupied and vacant parking spaces).
As part of the development process, strategy developers are pro-
vided with skeleton PDAs, i.e., PDAs that are fully functional and
lacking only their strategy layer. Each such agent possesses all the
functionalities it needs to interact with the simulation manager and
the simulated environment. This serves three purposes. First, it fa-
cilitates the strategy programming task as it allows people to focus
on this specific task, leaving out complex programming tasks such
as threading, communication, event handling, validation components,
etc. Second, it enables the programming of PDAs directly by people
with minimal programming background. Finally, the use of generic
functionality simplifies integration and debugging.
PARKING L OT S IMULATOR
The parking search problem typically considers a specific parking
lot design, including overall size, internal structure, number and loca-
tion of entry and exit points, alignment of different lanes, and permitted
driving directions. Different cars are constantly going into and out of
the parking lot. Every driver entering the parking lot has to individually
solve his parking problem by finding a free (valid) space in which to
park his car. The simulation can dictate the rate of cars arriving at the
parking lot during each time step and the rate of cars exiting it, and the
drivers’ observations of their vicinity in any location.
System’s Lower Bound
The performance evaluation of each of the new technologies was
based both on absolute performance and the magnitude of the relative
improvement obtained, measured in terms of costs. For the purpose
of calculating the relative improvement between any two information
models, a performance baseline (a “gold standard”) was required.
Obviously, the baseline is not zero since, as efficient as the agents can
be, there is still a lower theoretical limit to the system’s performance
(e.g„ whatever the theoretical allocation that maximizes the overall
social welfare may be, drivers still need to get to their parking spaces
and walk to the main entrance, reflecting some costs).
[attachment=60176]
Abstract
In this paper, we investigate the usefulness of peer-designed
agents (PDAs) as a turn-key technology for enhancing parking simulations.
The use of PDAs improves the system’s ability to capture the dynamics
of the interaction between individuals in the system, each theoretically
exhibiting a different strategic behavior. Furthermore, since people in
general are inherently rational and computation bounded, simulating this
domain becomes even more challenging. The advantage of PDAs in this
context lies in their ability to reliably simulate a large pool of human
individuals with diverse strategies and goals. We demonstrate the efficacy
of the proposed method by developing a large-scale simulation system
for the parking space search domain, which plays an important role in
urban transport systems. The system is based on 34 different parking
search strategies. Most of these strategies are substantially different from
synthetic strategies that are used in prior literature. A quantitative analysis
of the PDAs indicates that they reliably capture their designers’ real-life
strategies. Finally, we demonstrate the usefulness of PDA-based parking
space search simulation by utilizing it to evaluate four different informa-
tion technologies that are of increasing use in recent years.
INTRODUCTION
Simulation is an important tool for studying emergent behavior.
Using simulation, a researcher can carry out extensive data collection
in a simple, safe, and inexpensive way. While, in real transport
systems, the cost of a system evaluation is substantial, monetary, or
otherwise, the use of simulation enables evaluating various variants
of the system quickly and simply “with the click of a button.” Not
surprisingly, in recent years, agent technology in artificial intelligence
has become the dominating approach to developing simulation systems
and large-scale distributed systems [1]. While agents’ capabilities
substantially improve the modeling of the individuals that they repre-
sent, researchers have noted the challenge of constructing simulation
systems in domains where key players are people [2], particularly due
to their diverse behavior patterns.
RELATED WORK
Parking space search is an important problem for urban planners and
has been addressed by many traffic and transportation researchers [8],
[9]. Previous work in the area of parking simulations has placed little
emphasis on extracting user parking strategies, and most simulations
reported in the literature use a limited set of preprogrammed strategies.
For example, in PARKSIM, a network model is used to represent
the driver’s choice of behavior and consequently compare parking
facility layouts [10]. Other studies evaluate alternative parking system
designs and operating policies [11]. Recently, Jonkers et al. [9] have
recognized that people follow many different strategies when looking
for a parking space, yet their simulation focused on just two strategies.
To the best of our knowledge, no previous research in the area of
simulated parking lots incorporates a large collection of realistic
parking strategies as part of the entity design, as proposed in this paper.
P EER -D ESIGNED AGENT-BASED S IMULATION
The main idea in PDA-based design is to separate the logic of the
simulated agent’s strategy from the generic functionalities required to
have the agent participate in the simulation. This way, the strategy
development task can be distributed and delegated to a population that
is as similar as possible to the simulated population. We refer to these
people as strategy designers as they embed their strategies in PDAs.
The simulation is controlled and dynamic, and allows the use of
various agents of different types. Different simulation parameters can
be varied as well (e.g., using different parking lot structures, lanes,
and driving directions). The simulation also stores, at any given time,
the system’s current state (e.g., the location of cars and the location of
occupied and vacant parking spaces).
As part of the development process, strategy developers are pro-
vided with skeleton PDAs, i.e., PDAs that are fully functional and
lacking only their strategy layer. Each such agent possesses all the
functionalities it needs to interact with the simulation manager and
the simulated environment. This serves three purposes. First, it fa-
cilitates the strategy programming task as it allows people to focus
on this specific task, leaving out complex programming tasks such
as threading, communication, event handling, validation components,
etc. Second, it enables the programming of PDAs directly by people
with minimal programming background. Finally, the use of generic
functionality simplifies integration and debugging.
PARKING L OT S IMULATOR
The parking search problem typically considers a specific parking
lot design, including overall size, internal structure, number and loca-
tion of entry and exit points, alignment of different lanes, and permitted
driving directions. Different cars are constantly going into and out of
the parking lot. Every driver entering the parking lot has to individually
solve his parking problem by finding a free (valid) space in which to
park his car. The simulation can dictate the rate of cars arriving at the
parking lot during each time step and the rate of cars exiting it, and the
drivers’ observations of their vicinity in any location.
System’s Lower Bound
The performance evaluation of each of the new technologies was
based both on absolute performance and the magnitude of the relative
improvement obtained, measured in terms of costs. For the purpose
of calculating the relative improvement between any two information
models, a performance baseline (a “gold standard”) was required.
Obviously, the baseline is not zero since, as efficient as the agents can
be, there is still a lower theoretical limit to the system’s performance
(e.g„ whatever the theoretical allocation that maximizes the overall
social welfare may be, drivers still need to get to their parking spaces
and walk to the main entrance, reflecting some costs).