Our results show that with the Smart Rebalancing Policy,
we achieve a 51% improvement in taxi waiting time and
a 31% improvement in passenger waiting time over the
Observed Policy. Intuitively, we can explain the validity of
our results by considering a simple example of an airport
with two terminals, one with many taxis and no passengers
and the other with many passengers and no taxis. With the
Smart Rebalancing Policy, such situations are unlikely to
persist because the ChangiNOW server would immediately
send idle taxis from one terminal to pick up passengers from
the other, thereby creating a better matching of taxi supply
and demand so both taxis and passengers wait less. Our controlled
experiments used simulated taxi and passenger arrival
rates based on observed data. In actual implementation, we
believe similar results can be achieved by using both real
time taxi trajectories and ChangiNOW server requests in
our queuing model. Passenger arrival information in both
Our results show that with the Smart Rebalancing Policy,
we achieve a 51% improvement in taxi waiting time and
a 31% improvement in passenger waiting time over the
Observed Policy. Intuitively, we can explain the validity of
our results by considering a simple example of an airport
with two terminals, one with many taxis and no passengers
and the other with many passengers and no taxis. With the
Smart Rebalancing Policy, such situations are unlikely to
persist because the ChangiNOW server would immediately
send idle taxis from one terminal to pick up passengers from
the other, thereby creating a better matching of taxi supply
and demand so both taxis and passengers wait less. Our controlled
experiments used simulated taxi and passenger arrival
rates based on observed data. In actual implementation, we
believe similar results can be achieved by using both real
time taxi trajectories and ChangiNOW server requests in
our queuing model. Passenger arrival information in both