Airport Lounge Queuing Problem
Martin Ganco,
HyeJin Youn
1. Description
One hundred travelers are waiting in the passenger lounge of
the airport. When the airline agent at the gate announces that passengers may
begin to board the aircraft, people start forming the line at the gate. Our
model focuses on the line formation
mechanism, its impact on the distribution
of individual waiting times and evolution
of line preferences.
The key question that we ask is why,
when we assume that the objective of all passengers is to get inside the plane
as soon as possible, we observe that not all people rush to the line
immediately?
When we look at the dynamics in the waiting area more
carefully, there are usually some people who keep their seats until the queue
gets shorter. Some passengers stay seated while others tend to join the line
soon. Our model assigns a “line tolerance” parameter to every passenger at the
beginning of each boarding. Hence, a passenger's action - whether in the queue
or not - is determined by f(line
tolerance, number of people on the queue).
However, the overall waiting time is determined by the
ordering of agents as they enter the plane. The progression of line is slowed
down whenever passengers enter the plane in a “wrong” order – when a passenger
blocks seating of a passenger immediately behind. There is no explicit
coordination built into the model. The model was implemented in Matlab
and Python.
2. Assumptions
Agents:
- Tol (line tolerance): uniformly
distributed between [0:1], controls tendency to stay seated vs. join the
line (we tested the robustness of this assumption using Normal
distribution with no substantial impact on the shape of the curves and no
impact on our conclusions).
- ID (assigned
seat number): [1:100] represents a location of a seat from head to tail of
the cabin
- Action
(decision to join the line): bool, On - if agent decides to join the line, Off otherwise.
Environment:
- Nq: number of passengers in the queue
- The
cabin is assumed to be a long row of seats with numbers assigned in
ascending order
- The
seat number has no effect on the tendency of passengers to join the line
Procedure:
- T_delay: if agent ‘j’ has a lower ID than agent
'i' who is next in line, the agent 'i' has to wait for 'T_delay'.
- Fq: frequency of updating the queue by allowing
the first person to enter the plane
3. The Model
Initialization:
- Tol
is randomly given to every agent drawn from uniform distribution
- ID is given to every agent
from the interval [1:100]
Process:
- Agent i
is drawn randomly to determine whether it will join the queue or not
according to f(Tol, Nq). (The fundamental results are robust to different
matching methods, e.g. when the entire waiting area is searched for
passengers satisfying the rule in one time period, etc.)
- bool
f(Tol, Nq) = 1(Tol>Nq), 0 (otherwise)
- if f
is on, join the queue.
Note: the decision rule has the simple form: if g(line length) < line tolerance of
agent i => join the line, otherwise stay seated,
where in our case g(x)=x but g(.) can be any increasing function.
4. Results
4-1. No penalty for wrong order
- Hump
shaped line length over time
- Higher
mean line tolerance increases mean line length
- Uniformly
distributed individual waiting times (reflecting uniform preferences)
- Distribution
of line tolerance has no effect on the average waiting time (the mean
waiting time is the same due to exogenous line update frequency and no
penalty for wrong order)
-
Mean waiting time = 150
4-2. Penalty for wrong order
- Mean
waiting time increases with penalty
- The
line length decreases slowly with time
- Distribution
is no longer uniform - people with low line tolerance penalized disproportionately
more
- The
distribution seems bimodal (although, we did not perform significance
tests)
Mean waiting time = 450
5. Adding Evolution
As a simple extension, we ask the question whether the
passengers can evolve a line tolerance that would match their seating (if we assume
this is constant across runs). We add a simple hill climbing rule - each period
(airline boarding), each agent draws a new tolerance value. If it leads to
shorter time, it keeps it otherwise returns to the old tolerance value.
So, could the group evolve socially optimal line tolerance
that would "match" their seat assignment and optimize overall waiting
time?
The result shows that they cannot. After 100 runs the mean
waiting time and distribution did not change. The cause of the slowdown related
to the “wrong” order is a pure negative
externality. This result justifies an outside intervention.
6. Applications and Extensions
- In a general sense, the model
applies to situations where the objective is possible to achieve only when
an action is taken that yields temporary disutility (standing in line) and
status quo yields temporary benefits (keep sitting). The queue formation
model could be thus modified to model participation in negotiations,
perhaps social unrest, etc. The particular form of the model applies to
all similar line formation problems.
- As possible extensions, one
might consider agents traveling in groups like families or possible impact
of outside coordination by seat number groups or even individual line up
by IDs. However, in the current simple version of our model, the result of
such extensions seems obvious – decrease of mean waiting time and
convergence of distribution toward the uniform distribution.
7. Conclusion
The
purpose of our model was to investigate the mechanism of line formation and its
impact of the distribution of individual waiting times. Our results suggest
that delays cause a bimodal distribution of waiting times. The model also shows
that due to negative externality the decentralized agents do not have the
capability to optimize the overall boarding time by matching their line
tolerance with seat assignment. This indicates that outside intervention is
justified and necessary.