Abstract:
Mobility-on-demand (MoD) systems are emerging as a novel mode for urban mobility
which have enabled the users to have a reliable medium for mobility requirements.
Ride sharing services provided by these mobility on demand systems provide not just a
personalized experience in mobility but also present an immense potential for positive
societal impacts with reference to pollution, energy consumption, congestion, and etc.
Ride sharing services primarily concern with picking up spatiotemporally distributed
mobility demand and delivering it within a pre-specified time window subjected to different
other constraints. Large scale ride sharing in more sophisticated spatiotemporally
distributed mobility demand distributions require mathematical models and algorithms
in order to match riders and vehicle fleets in real time. As such common requirement
of models which address the potential of ride sharing is meticulously planned routes for
a fleet of taxis. Accordingly another such requirement is to incorporate meeting points
in route planning. In this work, we address the use of aforementioned two requirements
for ride sharing systems. Therefore our main contribution lies in the intersection of (1)
determining optimal routes for fleet of taxis with arbitrary set of constraints and, (2)
introducing meeting points to improve ride sharing.
We approach the optimal route planning problem by proposing two new types of
queries called Generalized Optimal Single Meeting Point Route (GOSMPR) query and
Generalized Optimal Multi Meeting Point Route (GOMMPR) query in road networks.
Given a vehicle current location 𝑠, a set of origin and destinations pairs of passenger
set 𝑃 and, an arbitrary set of constraints 𝐶, GOSMPR aims to find the optimal route
for the given taxi such that the selected route traverse all origins and destinations while
adhering to the constraints set C and minimizing the total path cost. In the same way,
GOMMPR address the same problem but by replacing each origin and destination with
set of pickup points and drop off points respectively. We show that problem of GOSMPR
and GOMMPR queries are NP-hard. Thus to address the GOSMPR problem efficiently,
we propose a dynamic programming approach inspired by 𝐴* algorithm. To reduce the
search state space of the proposed algorithm, we propose a series of state elimination
criteria and a novel variant of classical Held Karp lower bound of Travelling Salesman
Problem (TSP) as a heuristic. Also to address the GOMMPR problem, we propose
another dynamic programming based algorithm. We demonstrate that our proposed
iii
algorithm for GOSMPR can efficiently solve 10 passenger requests (21 points)(more
than the capacity of a typical ride sharing vehicle) optimally while the algorithm for
GOMMPR can solve up to 4 passenger requests within acceptable time limits.
We also conduct extensive experiments to demonstrate the impact of meeting points
in the context of ride sharing. We compare state of art ride sharing model RS-ILP and
our new meeting points based model RSMP-ILP to demonstrate the impact of meeting
points. We show that there is an improvement of 1.53% in service rates in RSMP-ILP
compared to RS-ILP which corresponds to 5048 trips in total.