Stochastic Route Planning For Electric Vehicles
Stochastic Route Planning For Electric Vehicles
Stochastic Route Planning For Electric Vehicles
Payas Rajan !
Department of Computer Science, University of California, Riverside, CA, USA
Chinya V. Ravishankar !
Department of Computer Science, University of California, Riverside, CA, USA
Abstract
Electric Vehicle routing is often modeled as a generalization of the energy-constrained shortest
path problem, taking travel times and energy consumptions on road network edges to be deter-
ministic. In practice, however, energy consumption and travel times are stochastic distributions,
typically estimated from real-world data. Consequently, real-world routing algorithms can make
only probabilistic feasibility guarantees. Current stochastic route planning methods either fail to
ensure that routes are energy-feasible, or if they do, have not been shown to scale well to large
graphs. Our work bridges this gap by finding routes to maximize on-time arrival probability and
the set of non-dominated routes under two criteria for stochastic route feasibility: E-feasibility and
p-feasibility. Our E-feasibility criterion ensures energy-feasibility in expectation, using expected
energy values along network edges. Our p-feasibility criterion accounts for the actual distribution
along edges, and keeps the stranding probability along the route below a user-specified threshold
p. We generalize the charging function propagation algorithm to accept stochastic edge weights to
find routes that maximize the probability of on-time arrival, while maintaining E- or p-feasibility.
We also extend multi-criteria Contraction Hierarchies to accept stochastic edge weights and offer
heuristics to speed up queries. Our experiments on a real-world road network instance of the Los
Angeles area show that our methods answer stochastic queries in reasonable time, that the two
criteria produce similar routes for longer deadlines, but that E-feasibility queries can be much faster
than p-feasibility queries.
Keywords and phrases Stochastic Routing, Electric Vehicles, Route Planning Algorithms
Acknowledgements The authors would like to thank the Mapbox Community team for access to
the Mapbox Traffic Data used in our experiments.
1 Introduction
Routing methods for Electric Vehicles (EVs) cannot just minimize travel time, but must also
address driver range anxiety. EVs have limited battery capacity, charging times are long,
and the charging infrastructure remains relatively sparse, so a major concern is stranding,
which occurs when the battery’s State of Charge (SoC) reaches zero en route. A route for an
EV is hence considered feasible only if the SoC along the route never reaches zero.
Merely trying to minimize travel time greatly increases the risk of stranding, since
energy consumption is typically quadratic in vehicle speed. Standard formulations such
as [9, 14, 30] model EV routing as a generalization of the NP-hard Constrained Shortest
Path problem [26, 58], and seek to minimize travel time while maintaining a non-zero SoC
along the route. Some recent work [7, 36] even tries to exercise direct control over travel
time and route feasibility, by pre-determining and assigning optimal EV travel speeds for
each road segment.
Most existing problem formulations also assume that travel times and energy consumption
values on road network edges are deterministic. In practice, both travel time and energy
consumption are stochastic, and difficult to estimate reliably [2, 17, 46]. In such a context,
even routing algorithms offering strong feasibility guarantees are of limited value. Approaches
that pre-determine and assign vehicle speeds for each edge are not practical, since speed is
not always a variable under driver control, but rather a result of prevalent traffic conditions.
Consequently, travel times and route feasibility may only be defined probabilistically.
Stochastic routing algorithms [12, 20, 40, 39, 38, 42, 43, 44, 27, 59], model travel times
along network edges as random variables with given probability distributions, and allow
richer query semantics, such as finding paths to maximize the probability of arrival before a
deadline [15], or finding the latest departure time and path to guarantee a certain probability
of arrival before a deadline [35]. Despite recent improvements [47], stochastic routing is
typically several orders of magnitude slower than deterministic routing, since obtaining travel
time distributions along a path requires very expensive convolutions of its edge distributions.
Limited work exists on stochastic route planning for EVs. Chen et al. [13] assume
lognormal travel-time and Gaussian energy-consumption distributions, and uses bicriteria
search to find the Pareto-optimal routes optimizing energy consumption and travel time
reliability. Jafari et al. [25] allow arbitrary distributions of travel times on edges and charging
stations, and uses multicriteria search to minimize the cost of charging and travel time, subject
to a minimum reliability threshold, on small synthetic graphs with randomly generated
edge weights and charging station placements. Shen et al. [51] allow correlated travel time
distributions between edges, and use bicriteria search on travel times and energy consumptions.
However, they assume deterministic energy consumptions, and run experiments on a network
of only a few hundred vertices.
E-Feasibility p-Feasibility
Non-Dominated Routes ✓ ✓
Probabilistic Budget Routes ✓ ✓
Stochastic
Routing
Figure 1 Stochastic route planning, classified by routing objective, edge distribution, and result.
Our work finds energy-feasible routes that maximize probability of arrival before deadline.
2 Related Work
EV routing has been typically modeled as energy-aware routing, with objective functions
ranging from minimizing the total energy consumption [14, 49], to minimizing travel time
while maintaining route feasibility [3, 9, 36, 54], to multicriteria search on both travel time and
energy consumption [22]. In contrast, most prior work on routing Internal Combustion-based
vehicles merely minimizes the total travel time [4, 53].
More attention is now being paid to real-world issues. Examples include allowing battery-
swapping stations [56], partial recharges at stations [9, 30, 8, 57] and maintaining battery
buffer to relieve range anxiety [45, 24, 18]. Many challenges remain, however. The energy
consumption models are imperfect, and factors such as battery wear, driver aggressiveness [31],
or traffic conditions are hard to model, but can have a significant impact. Data also suggests
that drivers may prefer familiar paths to shortest paths [60, 29].
By routing objective. Routing objectives can be quite varied, such as minimizing expected
time [11, 33, 32], maximizing the on-time-arrival probability [15], maintaining on-time-arrival
probability above a given threshold [35]. Some works [51, 13, 25] apply stochastic routing
algorithms to EVs, while others [1] route multiple EVs collaboratively, on-line.
By distribution. The edge distributions assumed can have a functional form, or be arbitrary
without a closed form. This choice also affects the edge weight representations used. For
functional forms, storing the distributional parameters suffices, but arbitrary distributions
require more space-intensive representations such as histograms. Further, with functional
forms, a small number of observations can suffice to capture real-world behaviour, but
histograms require much more data. Edge weight representations have been shown to
significantly affect the runtime performance of stochastic shortest path queries [44, 47].
SEA 2022
15:4 Stochastic Route Planning for Electric Vehicles
Output. Adaptive methods [41, 35] output policies for drivers to make routing decisions
on-line, as they reach vertices or edges during the drive. In contrast, a-priori approaches
produce routes before travel begins. [35] showed that adaptive approaches can produce
strictly better solutions than the a-priori approaches, but are much more computationally
expensive. Recently, performance improvements to policy-based approaches, such as the
Stochastic On-Time Arrival problem have also been proposed [48, 37, 27].
3 Problem Setup
A road network is a directed graph G = ⟨V, E⟩ where V is the set of vertices and E : V × V
is the set of edges. An s-t path P = [s = v1 , v2 · · · , vn = t] is a sequence of adjacent vertices
in the road network G. A set C ⊆ V is marked as charging stations.
▶ Definition 1 (State of Charge). The State of Charge (SoC) of an EV is the charge status
of the EV’s battery, lying between 0 and the battery capacity M . We denote the SoC on
arrival at a vertex v by v β and the SoC at departure from v by βv . We have βv ≥ v β if the
EV charges its batteries at node v, and βv = v β otherwise.
▶ Query 4 (Non-Dominated E-feasible Paths). Find the set of E-feasible s-t paths such that
their travel time distributions are not dominated by any other path.
▶ Query 5 (Probabilistic Budget E-feasible Path). Find an E-feasible s-t path that maximizes
the probability of reaching t before a given deadline d.
▶ Query 7 (Non-Dominated p-Feasible Paths). Find the set of s-t paths whose travel time
distributions are not dominated by any other path, and which ensure that probability not being
stranded is at least p.
▶ Query 8 (Probabilistic Budget p-Feasible Paths). Find an s-t path which maximizes the
probability of reaching t before a given deadline d, while keeping the probability of not being
stranded is at least p.
SEA 2022
15:6 Stochastic Route Planning for Electric Vehicles
SoC SoC
Probability
Probability
Probability
Probability
Probability
0.4 0.4 0.4 0.4 0.4
Figure 2 E-feasible queries. Edges have two weights: a travel time distribution (below), and
an expected energy depletion (above). Shaded nodes are charging stations, with piecewise-linear
charging functions. The CFP search propagates travel time distributions using convolutions.
However, complexities arise since we use stochastic travel times. The deterministic case
can simply use a min-priority queue ordered by travel times, but distributions can be ordered
in different ways. For simplicity, we will use usual stochastic ordering [50] to order the travel
time distributions in the priority queue, under which two random variables X and Y obey
X ⪯ Y iff Pr[X > x] ≤ Pr[Y > x], ∀x ∈ (−∞, ∞). Other stochastic orderings, such as the
hazard rate or likelihood ratio ordering, may result in interesting tradeoffs for the EV, but are
beyond the scope of this paper. Also, deterministic travel times can be simply added along a
path, but travel time distributions must be convolved to aggregate travel time distributions.
For expected-feasible routes, we will use stochastic travel times, but expected values for
energy depletion. That is, let e1 , e2 , . . . , en be the edges comprising a path P , and let edge ei
have travel time and energy depletion distributions Ti and Di . For expected-feasible routing,
the aggregate travel time distribution TP = T1 ∗ T2 ∗ · · · ∗ Tn , and the aggregate energy
depletion value is E[DP ] = E[D1 ] + E[D2 ] + . . . + E[Dn ].
The depletion function for a null path comprising a single vertex s is the identity depletion
function δ∅ : βs 7→ βs . Let P1 = [vi , vi+1 , . . . , vj ] and P2 = [vj+1 , vj+2 , . . . , vk ] be contiguous
segments, and P = P1 P2 = [vi , . . . , vk ] be their concatenation. In this case, we have
sP = max{sP1 , cP1 + sP2 }, eP = min{eP2 , eP1 − cP2 } and cP = cP1 + cP2 .
P. Rajan and C. V. Ravishankar 15:7
Since the charging function Φc (·) is assumed to be piecewise linear, its breakpoints induce
breakpoints for bℓ . For a given value of τ we need to create one label per breakpoint of
bℓ [8]. For a fixed τ and each breakpoint B = (tB , SoCB ) of bℓ , we add to Lu (v) the label
⟨tB , SoCB , v, T∅ ⟩, and update the travel times to v in QG .
SEA 2022
15:8 Stochastic Route Planning for Electric Vehicles
SoC SoC
s t
0.6 0.6 0.6 0.6 0.6
Probability
Probability
Probability
Probability
Probability
0.4 0.4 0.4 0.4 0.4
0.6
Probability
Probability
0.6 0.75
Probability
Probability
0.75 0.75
Figure 3 p-Feasible queries. Travel time and energy depletion are both distributions., propagated
by the CFP search using convolutions. While the non-dominated search stops only when QG becomes
empty, the probabilistic budget route search can stop when TP (t) drops to 0.
4. At the destination t: End search, backtrack using parent pointers to extract an s-t path.
A label ℓ is said to dominate another label ℓ′ if bℓ (t, τ ) ≥ bℓ′ (t, τ ) for all t > 0 and all τ > 0.
If we end the search only when QG is empty, not simply when t is reached, we obtain the
E-feasible non-dominated paths. For E-feasible probabilistic budget paths, we end the search
when TP (d) = 0, i.e. the probability of reaching t within the time budget d drops to 0.
For p-feasible routing, we must consider the actual depletion distribution DP for a path
P , not simply E[DP ], which sufficed for expected-feasible paths. As with expected-feasible
paths, we must also deal with the travel time distribution TP . If path P has the edges
e1 , e2 , . . . , en , then TP = T1 ∗ T2 ∗ · · · ∗ Tn and DP = D1 ∗ D2 ∗ · · · ∗ Dn . We can use p to
place a bound on the maximum energy depletion we can accommodate over a path P . Let
so that cP (p) is the highest energy depletion that could occur along P with a probability of
no more than p, that is, to ensure a non-stranding probability of p.
P. Rajan and C. V. Ravishankar 15:9
Let σ∅ be the identity stochastic depletion profile for a null path, so that σ∅ (βs , p) = βs . If
P1 = [vi , vi+1 , . . . , vj ] and P2 = [vj+1 , vj+2 , . . . , vk ], the depletion profile of the concatenated
path P = P1 P2 = [vi . . . vk ] is given by sP (p) = max{sP1 (p), cP1 (p) + sP2 (p)}, eP (p) =
min{eP2 (p), eP1 (p) − cP2 (p)} and cP = cP1 ∗ cP2 .
Since Φc (·) is piecewise linear, its breakpoints induce breakpoints for bℓ . Moreover, p is
already known at query time, so for a given value of τ , we only need to create one label
per breakpoint of b′ℓ [8]. For a fixed τ and each breakpoint B = (tB , SoCB ) of b′ℓ , we add
to Lu (v) the label ⟨tB , SoCB , v, T∅ ⟩, and update the travel times to v in QG .
As with E-feasible routes, τ can take an infinite number of values, but we use histograms
to represent Te , and we need to generate only one set of breakpoints per histogram bin.
Lastly, we verify p-feasibility of path [s . . . v], by maintaining the running product of the
non-stranding probabilities of all legs over this path. If this product falls below p, the
path [s . . . v] is no longer p-feasible. The search is pruned and labels for v are dropped.
SEA 2022
15:10 Stochastic Route Planning for Electric Vehicles
4. At the destination t: End search, backtrack using parent pointers to extract an s-t path.
For a given p, a label ℓ dominates another label ℓ′ if b′ℓ (t, τ, p) ≥ b′ℓ (t, τ, p) for all t > 0 and
τ > 0. If the search terminates only when QG is empty, the resulting s-t paths are the
p-feasible Non-Dominated Paths. For Probabilistic Budget queries, we end the search only
when it reaches far enough for the probability of reaching t within the time budget d is 0.
7 Experiments
Our algorithms were implemented in Rust 1.60.0-nightly with full optimizations and run on
an Intel core i5-8600K processor with 3.6GHz base clock, 192KB of L1, 1.5 MB of L2, and 9
MB of L3 cache and equipped with 64GB of dual-channel 3200MHz DDR4 RAM.
1
https://www.mapbox.com/traffic-data
2
https://labs.mapbox.com/what-the-tile
P. Rajan and C. V. Ravishankar 15:11
into weekday and weekend speed histograms. We discarded the weekend histograms due to
sparsity, and used only the weekday speeds for our experiments. We added latitudes and
longitudes for each vertex from the OSM dataset taken from GeoFabrik,3 contracted the
degree-2 vertices, and extracted the largest connected component. This step resulted in the
final routing graph of 244,728 vertices and 453,942 edges.
We added elevation data from the NASADEM dataset [34] at 30M resolution to each
vertex, using bilinear interpolation to estimate elevations at vertex locations. Lastly, we
obtained charging stations from the Alternative Fuels Data Center,4 marking the vertex
closest to each charging station as the charging vertex. The charging function Φc on each
vertex c was linear, and either (1) a slow, charging to 100% in 100 minutes, or (2) fast,
charging up to 80% in 30 minutes, and up to 100% in 60 minutes. We randomly assigned the
slow charging function to 70% of charging stations, the fast charging function to the rest.
Energy consumption parameters for εe on all edges e were derived using the vertex
elevations and the values ae , be , ce , de used for Nissan Leaf 2013 in [16]. To force the search
to require charging en route for feasibility, we assumed that the EV had a 12 kWh battery.
7.2 Results
Using stochastic edge weights raises many challenges that do not arise for deterministic weights.
Two obvious issues are maintaining route feasibility, and aggregating edge distributions Te
and De into path distributions TP or DP , which requires expensive convolutions. Several
other issues also arise, two of which we will discuss.
Number of histogram bins. The time and energy value ranges in the path distributions
TP , DP increases linearly with the number of edges in P , so more histogram bins are needed
to maintain accuracy. As in the deterministic case, the Dijkstra search labels track the travel
time-charging time tradeoff. The labels represent histograms, so the label sizes increase with
the number bins used for TP and DP . Labels become progressively larger for longer routes,
raising the cost of all operations on the distributions, (convolution, dominance checks, etc.).
3
https://download.geofabrik.de/north-america/us/california/socal.html
4
https://afdc.energy.gov/fuels/electricity_locations.html
SEA 2022
15:12 Stochastic Route Planning for Electric Vehicles
At charging stations, moreover, we must create a set of breakpoints per bin of the energy
depletion histogram. More breakpoints are created for charging stations further along the
route, increasing costs and making label dominance checks labels more difficult.
We also note that the CH shortcuts represent longer routes, whose histograms have more
bins than the original graph edges. Shortcut edges are hence more expensive to handle than
original graph edges, decreasing the utility of shortcuts in speeding up route planning queries.
Ensuring stochastic feasibility. Standard probabilistic budget routes use a single criterion,
such as travel time [42, 47]. In contrast, our queries must handle search with two criteria to
maintain feasibility. Further, the number of breakpoints in the charging functions along a
route determines the number of labels generated.
For deterministic edge weights, path costs are just sums of edge costs, so routing takes
just microseconds even on continent-sized road networks [4]. Routing with stochastic edge
weights is far slower, since the convolutions needed to get path costs are very expensive.
Prior work [42, 47] deals only with stochasticity in time, ignoring energy feasibility, but we
consider both aspects. Our methods take tens of seconds, which is comparable to these prior
methods. In preliminary experiments, our use of stochastic, multicriteria CH yielded a 2–2.4
factor speedup over queries not using CH. In deterministic settings, similar methods have
been reported to achieve speedups of two to three orders of magnitude [21]. This lower gain
can be attributed to the weaker “hierarchy” with stochastic edge weights, causing far more
shortcuts to be added to the original graph. This forces the Dijkstra search to scan many
more edges on settling each vertex.
Table 2 Single-criterion probabilistic budget routing queries [47] vs. our E-feasible and p-feasible
queries on the Tile 0230123 graph. Query times (seconds) are averages over 100 random vertex pairs.
The EV is a Nissan Leaf 2013 with 12 kWh battery and 50% starting SoC.
Table 3 E-feasible and p-feasible query performance on the Tile 0230123 graph, with real-world
charging station and elevation data. Query times (seconds) are over 500 random vertex pairs. EV
used is a Nissan Leaf 2013 fitted with a 12 kWh battery and 50% starting SoC.
Table 4 Average Jaccard Index for 500 random E-feasible and p-feasible routes, with p = 0.85.
The index is 0 when the routes are edge-disjoint, and 1 when they are identical.
Table 4 shows how similar the E-feasible and p-feasible routes are, using the average
Jaccard Similarity between the set edges of a route chosen by each of them. The Jaccard
similarity for two routes P1 and P2 is the number of edges common to both divided by the
number of edges in their union. That is,
|{e ∈ P1 } ∩ {e ∈ P2 }|
J(P1 , P2 ) =
|{e ∈ P1 } ∪ {e ∈ P2 }|
The Jaccard index clearly increases with the time budget, so the E-feasible and p-feasible
routes are more similar when the routes are longer. This is because longer routes require
more convolutions, making DP closer to the Gaussian, which is more concentrated near its
mean. In such cases, the pruning of edges forced by the feasibility criterion brings the set
of edges of E-feasible routes closer to the set of edges for p-feasible routing. For shorter
routes, however, the difference between the two types of queries is higher. Hence, if stronger
feasibility guarantees are desired for shorter routes, p-feasible queries may be better.
SEA 2022
15:14 Stochastic Route Planning for Electric Vehicles
stochastic edge weights, while allowing recharging stations. We also applied a multicriteria
variant of stochastic Contraction Hierarchies to speed up our queries, using the restricted
stochastic dominance criterion of [10] and the ϵ-dominance among labels. We demonstrated
that our techniques were feasible in the real world by running experiments on a realistic
routing instance, using real-world travel speeds in the Los Angeles area collected over four
and a half months. The similarity between E-feasible and p-feasible routes indicates the
potential applicability of a tiered-hierarchy style approach [47] to help speed up stochastic
feasible routing queries even further, and could be an interesting avenue for further work.
References
1 Niklas Akerblöm, Yuxin Chen, and Morteza Haghir Chehreghani. An online learning framework
for Energy-Efficient navigation of electric vehicles. In Proceedings of the Twenty-Ninth
International Joint Conference on Artificial Intelligence (IJCAI-20), pages 2051–2057, 2020.
2 Yazan Al-Wreikat, Clara Serrano, and José Ricardo Sodré. Driving behaviour and trip
condition effects on the energy consumption of an electric vehicle under real-world driving.
Appl. Energy, 297:117096, September 2021.
3 Andreas Artmeier, Julian Haselmayr, Martin Leucker, and Martin Sachenbacher. The shortest
path problem revisited: Optimal routing for electric vehicles. In KI 2010: Advances in Artificial
Intelligence, Lecture Notes in Computer Science, pages 309–316. Springer, Berlin, Heidelberg,
September 2010.
4 Hannah Bast, Daniel Delling, Andrew Goldberg, Matthias Müller-Hannemann, Thomas Pajor,
Peter Sanders, Dorothea Wagner, and Renato F Werneck. Route planning in transportation
networks. In Algorithm Engineering, Lecture Notes in Computer Science, pages 19–80. Springer,
Cham, 2016.
5 Lucas S Batista, Felipe Campelo, Frederico G Guimarães, and Jaime A Ramírez. A comparison
of dominance criteria in many-objective optimization problems. In 2011 IEEE Congress of
Evolutionary Computation (CEC), pages 2359–2366, June 2011.
6 Reinhard Bauer, Tobias Columbus, Ignaz Rutter, and Dorothea Wagner. Search-space size in
contraction hierarchies. Theor. Comput. Sci., 645:112–127, September 2016.
7 Moritz Baum. Engineering Route Planning Algorithms for Battery Electric Vehicles. PhD
thesis, Karlsruhe Institute of Technology, 2018.
8 Moritz Baum, Julian Dibbelt, Andreas Gemsa, Dorothea Wagner, and Tobias Zündorf. Shortest
feasible paths with charging stops for battery electric vehicles. In Proceedings of the 23rd
SIGSPATIAL International Conference on Advances in Geographic Information Systems,
page 44. ACM, November 2015.
9 Moritz Baum, Julian Dibbelt, Andreas Gemsa, Dorothea Wagner, and Tobias Zündorf. Shortest
feasible paths with charging stops for battery electric vehicles. Transportation Science,
53(6):1627–1655, November 2019.
10 Roger L Berger. A nonparametric, intersection-union test for stochastic order. Technical
report, North Carolina State University. Dept. of Statistics, 1986.
11 Dimitri P Bertsekas and John N Tsitsiklis. An analysis of stochastic shortest path problems.
Math. Oper. Res., 16(3):580–595, August 1991.
12 Anthony Chen and Zhaowang Ji. Path finding under uncertainty. J. Adv. Transp., 39(1):19–37,
September 2005.
13 Xiao-Wei Chen, Bi Yu Chen, William H K Lam, Mei Lam Tam, and Wei Ma. A bi-objective re-
liable path-finding algorithm for battery electric vehicle routing. Expert Syst. Appl., 182:115228,
November 2021.
14 Jochen Eisner, Stefan Funke, and Sabine Storandt. Optimal route planning for electric vehicles
in large networks. In AAAI, 2011.
15 Y Y Fan, R E Kalaba, and J E Moore. Arriving on time. J. Optim. Theory Appl., 127(3):497–
513, December 2005.
P. Rajan and C. V. Ravishankar 15:15
16 Chiara Fiori, Kyoungho Ahn, and Hesham A Rakha. Power-based electric vehicle energy
consumption model: Model development and validation. Appl. Energy, 168:257–268, April
2016.
17 Chiara Fiori, Vittorio Marzano, Vincenzo Punzo, and Marcello Montanino. Energy consumption
modeling in presence of uncertainty. IEEE Trans. Intell. Transp. Syst., pages 1–12, 2020.
18 Matthew William Fontana. Optimal routes for electric vehicles facing uncertainty, congestion,
and energy constraints. PhD thesis, Massachusetts Institute of Technology, 2013.
19 Mogens Fosgerau. The valuation of travel time variability. Technical report, International
Transport Forum, 2016.
20 H Frank. Shortest paths in probabilistic graphs. Oper. Res., 17(4):583–599, 1969.
21 Robert Geisberger, Peter Sanders, Dominik Schultes, and Daniel Delling. Contraction hierar-
chies: Faster and simpler hierarchical routing in road networks. In Experimental Algorithms,
pages 319–333. Springer, Berlin, Heidelberg, May 2008.
22 Michael T Goodrich and Paweł Pszona. Two-phase bicriterion search for finding fast and
efficient electric vehicle routes. In Proceedings of the 22nd ACM SIGSPATIAL International
Conference on Advances in Geographic Information Systems, pages 193–202. ACM, November
2014.
23 Ming Hua and Jian Pei. Probabilistic path queries in road networks: traffic uncertainty aware
path selection. In Proceedings of the 13th International Conference on Extending Database
Technology, EDBT ’10, pages 347–358, New York, NY, USA, March 2010. Association for
Computing Machinery.
24 Gerhard Huber, Klaus Bogenberger, and Hans van Lint. Optimization of charging strategies
for battery electric vehicles under uncertainty. IEEE Trans. Intell. Transp. Syst., pages 1–17,
2020.
25 Ehsan Jafari and Stephen D Boyles. Multicriteria stochastic shortest path problem for electric
vehicles. Networks Spat. Econ., 17(3):1043–1070, September 2017.
26 H C Joksch. The shortest route problem with constraints. J. Math. Anal. Appl., 14(2):191–197,
May 1966.
27 Moritz Kobitzsch, Samitha Samaranayake, and Dennis Schieferdecker. Pruning techniques
for the stochastic on-time arrival problem- an experimental study. arXiv, July 2014. arXiv:
1407.8295.
28 Sejoon Lim, Christian Sommer, Evdokia Nikolova, and Daniela Rus. Practical route planning
under delay uncertainty: Stochastic shortest path queries. In Robotics: Science and Systems,
volume 8, pages 249–256. books.google.com, 2013.
29 Antonio Lima, Rade Stanojevic, Dina Papagiannaki, Pablo Rodriguez, and Marta C González.
Understanding individual routing behaviour. J. R. Soc. Interface, 13(116), March 2016.
30 Sören Merting, Christian Schwan, and Martin Strehler. Routing of electric vehicles: Con-
strained shortest path problems with resource recovering nodes. In OASIcs-OpenAccess
Series in Informatics, volume 48. Schloss Dagstuhl - Leibniz-Zentrum fuer Informatik GmbH,
Wadern/Saarbruecken, Germany, 2015.
31 Sebastiano Milardo, Punit Rathore, Marco Amorim, Umberto Fugiglando, Paolo Santi, and
Carlo Ratti. Understanding drivers’ stress and interactions with vehicle systems through
naturalistic data analysis. IEEE Trans. Intell. Transp. Syst., pages 1–12, 2021.
32 Elise D Miller-Hooks and Hani S Mahmassani. Least expected time paths in stochastic,
Time-Varying transportation networks. Transportation Science, 34(2):198–215, May 2000.
33 Elise Deborah Miller-Hooks. Optimal routing in time-varying, stochastic networks: Algorithms
and implementations. PhD thesis, The University of Texas at Austin, 1997.
34 J P L Nasa. NASADEM merged DEM global 1 arc second V001 [dataset], 2020. Accessed:
2021-6-9. doi:10.5067/MEaSUREs/NASADEM/NASADEM_HGT.001.
35 Yu (marco) Nie and Xing Wu. Shortest path problem considering on-time arrival probability.
Trans. Res. Part B: Methodol., 43(6):597–613, July 2009.
SEA 2022
15:16 Stochastic Route Planning for Electric Vehicles
36 Patrick Niklaus. A unified framework for electric vehicle routing. Master’s thesis, Karlsruhe
Institute of Technology, 2017.
37 Mehrdad Niknami and Samitha Samaranayake. Tractable pathfinding for the stochastic
On-Time arrival problem. In Experimental Algorithms, pages 231–245. Springer International
Publishing, 2016.
38 Evdokia Nikolova. Approximation algorithms for reliable stochastic combinatorial optimiza-
tion. In Approximation, Randomization, and Combinatorial Optimization. Algorithms and
Techniques, pages 338–351. Springer Berlin Heidelberg, 2010.
39 Evdokia Nikolova, Matthew Brand, and David R Karger. Optimal route planning under
uncertainty. ICAPS, 2006.
40 Evdokia Nikolova, Jonathan A Kelner, Matthew Brand, and Michael Mitzenmacher. Stochastic
shortest paths via quasi-convex maximization. In Algorithms – ESA 2006, pages 552–563.
Springer Berlin Heidelberg, 2006.
41 Axel Parmentier and Frédéric Meunier. Stochastic shortest paths and risk measures. arXiv,
August 2014. arXiv:1408.0272.
42 Simon Aagaard Pedersen, Bin Yang, and Christian S Jensen. Fast stochastic routing under
time-varying uncertainty. VLDB J., October 2019.
43 Simon Aagaard Pedersen, Bin Yang, and Christian S Jensen. Anytime stochastic routing with
hybrid learning. Proceedings VLDB Endowment, 13(9):1555–1567, May 2020.
44 Simon Aagaard Pedersen, Bin Yang, and Christian S Jensen. A hybrid learning approach to
stochastic routing. In 2020 IEEE 36th International Conference on Data Engineering (ICDE),
pages 1910–1913, April 2020.
45 Payas Rajan, Moritz Baum, Michael Wegner, Tobias Zündorf, Christian J West, Dennis
Schieferdecker, and Daniel Delling. Robustness generalizations of the shortest feasible path
problem for electric vehicles. In Matthias And Federico, Müller-Hannemann, editor, Proceedings
of 21st Symposium on Algorithmic Approaches for Transportation Modelling, Optimization,
and Systems (ATMOS 2021), Open Access Series in Informatics (OASIcs), pages 11:1–11:18,
Dagstuhl, Germany, September 2021. Schloss Dagstuhl - Leibniz-Zentrum für Informatik.
46 Payas Rajan and Chinya V Ravishankar. The phase abstraction for estimating energy
consumption and travel times for electric vehicle route planning. In Proceedings of the 27th
ACM SIGSPATIAL International Conference on Advances in Geographic Information Systems,
SIGSPATIAL ’19, pages 556–559, New York, NY, USA, 2019. ACM.
47 Payas Rajan and Chinya V Ravishankar. Tiering in contraction and edge hierarchies for
stochastic route planning. In Proceedings of the 29th International Conference on Advances
in Geographic Information Systems, SIGSPATIAL ’21, pages 616–625, New York, NY, USA,
November 2021. Association for Computing Machinery.
48 Guillaume Sabran, Samitha Samaranayake, and Alexandre Bayen. Precomputation techniques
for the stochastic on-time arrival problem. In 2014 Proceedings of the Meeting on Algorithm
Engineering and Experiments (ALENEX), Proceedings, pages 138–146. Society for Industrial
and Applied Mathematics, December 2013.
49 M Sachenbacher, M Leucker, A Artmeier, and J Haselmayr. Efficient Energy-Optimal routing
for electric vehicles. AAAI, 2011.
50 Moshe Shaked and J George Shanthikumar. Stochastic Orders. Springer New York, October
2006.
51 Liang Shen, Hu Shao, Ting Wu, William H K Lam, and Emily C Zhu. An energy-efficient
reliable path finding algorithm for stochastic road networks with electric vehicles. Transp. Res.
Part C: Emerg. Technol., 102:450–473, May 2019.
52 Kenneth A Small, Clifford Winston, and Jia Yan. Uncovering the distribution of motorists’
preferences for travel time and reliability. Econometrica, 73(4):1367–1382, July 2005.
53 Christian Sommer. Shortest-path queries in static networks. ACM Computing Surveys (CSUR),
46(4):45, April 2014.
P. Rajan and C. V. Ravishankar 15:17
54 Sabine Storandt. Quick and energy-efficient routes: Computing constrained shortest paths for
electric vehicles. In Proceedings of the 5th ACM SIGSPATIAL International Workshop on
Computational Transportation Science, IWCTS ’12, pages 20–25, New York, NY, USA, 2012.
ACM.
55 Sabine Storandt. Route planning for bicycles – exact constrained shortest paths made practical
via contraction hierarchy. In Proceedings of the Twenty-Second International Conference on
Automated Planning and Scheduling, 2012.
56 Sabine Storandt and Stefan Funke. Cruising with a Battery-Powered vehicle and not getting
stranded. In AAAI, volume 3, page 46, 2012.
57 Martin Strehler, Sören Merting, and Christian Schwan. Energy-efficient shortest routes for
electric and hybrid vehicles. Trans. Res. Part B: Methodol., 103(Supplement C):111–135,
September 2017.
58 Christoph Witzgall and Alan J Goldman. Most profitable routing before maintenance. In
OPERATIONS RESEARCH, page B82. INST OPERATIONS RESEARCH MANAGEMENT
SCIENCES 901 ELKRIDGE LANDING RD, STE 400, LINTHICUM HTS, MD, USA, 1965.
59 Bin Yang, Chenjuan Guo, Christian S Jensen, Manohar Kaul, and Shuo Shang. Stochastic
skyline route planning under time-varying uncertainty. In 2014 IEEE 30th International
Conference on Data Engineering, pages 136–147, March 2014.
60 Shanjiang Zhu and David Levinson. Do people use the shortest path? an empirical test of
wardrop’s first principle. PLoS One, 10(8):e0134322, August 2015.
61 Weiwei Zhuang, Yadong Li, and Guoxin Qiu. Statistical inference for a relaxation index of
stochastic dominance under density ratio model. J. Appl. Stat., pages 1–19, August 2021.
SEA 2022