2018 IEEE International Conference on Vehicular Electronics and Safety (ICVES)

September 12-14, 2018, Madrid, Spain

Smooth Obstacle Avoidance Path Planning for

Autonomous Vehicles
Wael Ben-Messaoud Michel Basset Jean-Philippe Lauffenburger Rodolfo Orjuela
IRIMAS EA7499 IRIMAS EA7499 IRIMAS EA7499 IRIMAS EA7499
Université de Haute-Alsace Université de Haute-Alsace Université de Haute-Alsace Université de Haute-Alsace
Mulhouse, France Mulhouse, France Mulhouse, France Mulhouse, France
wael.ben-messaoud@uha.fr michel.basset@uha.fr jean-philippe.lauffenburger@uha.fr rodolfo.orjuela@uha.fr

Abstract—Obstacle avoidance and overtaking are important map and a localization system). The most common used
manoeuvres for autonomous driving. The collision avoidance with methods for global path planning are Dijkstra [7], A∗ and
a vehicle, a pedestrian or any obstacle and the generation of Rapidly-exploring Random Tree RRT [8]. These methods are
a feasible continuous curvature trajectory represent the major
problems faced by researchers to provide a safe path planning time-consuming as the calculation of the whole trajectory in
solution. This paper presents an algorithm able to avoid vehicles real time is slow especially when the map is wide and thus, it
or obstacles by proposing a smooth local modified trajectory of can hardly be applied in real time especially for autonomous
a global path. The proposed method is based on the combination cars and when obstacles are dynamic.
of a parameterized sigmoid function and a rolling horizon. Local approaches restrict the trajectory calculation only
The major advantage of this method is the reactivity to the
obstacle motion and the generation of a smooth trajectory in on a limited time/distance window by taking the perceived
a low execution time. The lateral and the longitudinal safety environment into account. The local path is limited to the
distances are easily parameterized when generating the avoidance perception range of the sensors or camera. These methods are
trajectory. Simulation results show that the algorithm performs used as the whole path is unknown in the majority of cases.
collision avoidance manoeuvres for static and dynamic obstacles The reference trajectory is calculated simultaneously when the
in an effective way. The method is validated by applying the
avoidance approach on real measured trajectory. displacement is started. Using these methods, the navigation
Index Terms—autonomous vehicle, path planning, obstacle can be ensured in a partially known environment and also with
avoidance, sigmoid function. dynamic obstacles.
Several local methods based on interpolating the trajectory
I. INTRODUCTION by mathematical expressions are proposed in the literature.
Autonomous driving is currently an active research sub- Clothoid curves [9], [10], polynomial curves [11], Bézier
ject especially for perception and collision avoidance. An curves and spline curves [12], [13] are some interesting
autonomously driven vehicle must, first of all, be able to examples of these methods. Clothoid curves are used for
determine its path with respect to the perceived static and instance in tentacle methods [10]. The latter is based on the
dynamic environment. After that, it plans the movement to selection of one trajectory from a set of candidate trajectories
be carried out to achieve its objective by respecting a certain using an optimization criterion. Yet, its major issue is the
number of constraints related to the environment (presence long computation time because a great number of paths
of obstacles) and vehicle dynamics (planning stage). Thus, is generated at each step before the selection of the best
the complete process that allows a vehicle to navigate on candidate through optimization. Polynomial, Bézier and Spline
its environment can be summed up in highly interdependent curves are able to generate a smooth trajectory by linking
modules: perception (including localization), planning and different way-points representing the free-space navigation for
control [1]. Trajectory planning for autonomous vehicles has the vehicle. A better compromise must be reached between
been studied in different fields like robotics, autonomous cars, the order of the mathematical function and the computation
unmanned vehicles, underwater vehicles and drones [2]–[6]. time. Sigmoid-based curve is a local planning method based
The main challenge when generating a trajectory is ensuring on an easily configurable path [14]. The parameters of the
its feasibility by taking into consideration several constraints developed sigmoid function are able to adjust the smoothness
related to vehicle dynamics, the road profile, etc. The need of the obstacle avoidance manoeuvre and the safety distance
of a low execution time is also an essential factor in order between the ego vehicle and the leading vehicle to be avoided.
to be implemented in real time. We can distinguish two In local planning, planning area or horizon computation is
categories of path planning approach for obstacle avoidance: performed such as in the Horizon Planning Approach which
the global planning and the local planning. The principle consists in planning the trajectory by dividing the space into
of global approaches is to determine a complete movement regions [15], [16]. This allows quickly computing trajectories
between an initial position and a final position from a priori by dividing the problem into sub-problems. The horizon can
known environment information (e.g. provided by a digital be expressed in time [17] or in distance [18], [19].

978-1-5386-3543-8/18/$31.00 ©2018 IEEE

 In this paper, we will present a new algorithm based on vehicle and the obstacles. Long range sensors like Light
the sigmoid function and the rolling horizon method. The Detection and Ranging (LIDAR) sensors and camera can for
proposed approach is able to locally plan trajectories for instance be used to detect and localize obstacles [20], [21]. The
autonomous vehicles, especially for collision avoidance and dimensions of the avoided obstacle or vehicle are supposed to
overtaking manoeuvers. Based on a global reference path, be known. In practice, the dimensions of the avoided obstacle
the algorithm generates a local trajectory in a very small can be measured in real time using LIDAR. We also assume
execution time (less that a few tens of µs as illustrated that the moving obstacle velocities are known and that the
in the simulation section) to avoid the static and dynamic avoidance task is performed with constant longitudinal speed
obstacles. The smooth path in the rolling horizon is optimized for the ego-vehicle.
in terms of the execution time when calculating the safe Our developed work associates two approaches: [14] where
obstacle avoidance trajectory. The paper is decomposed into the authors propose the use of a sigmoid function to avoid
three sections: in the second section, the problem statement the obstacle and the use of a rolling horizon to plan for a
of the obstacle avoidance methods is defined. In section III, decomposed area of the path [15]. We propose to combine
the new proposed method is explained and the steps of the these ideas to provide an easily configurable continuous cur-
algorithm are presented. Finally, the validation of the method vature path, updated at every step and to guarantee a lateral
is done using simulation results for different situations and for and a longitudinal safety distance to avoid obstacles. The
real measured trajectory. rolling horizon approach will allow to decompose the global
trajectory to local parts and calculate at each horizon step the
II. PROBLEM STATEMENT next predicted trajectory.
The main issue when avoiding detected obstacles or over-
A. Sigmoid-based function
taking vehicles is to propose an easily configurable online
method to modify an initial trajectory. The challenge is to find The sigmoid function (1) proposed in [14] is used to add a
a compromise between the expected configurable trajectory lateral offset from a straight line.
and its reactivity to the variation of the obstacles position. B
The easy configuration of the path is necessary to correlate y(x) = (1)
1 + e(−a(x−c))
it to the vehicle speed. The low execution time of the gen-
erated trajectory represents an important criterion also as the where y(x) is the lateral offset of the vehicle and x is the po-
trajectory must be implemented in real time. sition in longitudinal direction; B represents the way position
This paper deals with a local path modification of a global P3 to reach to generate the obstacle avoidance manoeuvre; c
reference trajectory. Its focus is to propose a safe trajectory modifies the shape of the function and a is the slope of the
by avoiding static and dynamic obstacles thanks to a lane sigmoid. Fig. 2 illustrates the shape of the function and its
change manoeuvre. The initial trajectory can be provided curvature when considering three values of the parameter a.
from a global path planner using for instance a digital map Noticing that the point P2 is independent to a and its position
database or a recorded map. The architecture to control the can only be imposed by c.
autonomous vehicle is generally composed of different steps
(see Fig. 1) which are perception (including localization), P3
trajectory generation and control such as the one developed
in our laboratory [1]. P2

P1

Fig. 2. Illustration of the sigmoid curve proposed in [22].

The inputs of the function are the obstacle position and the
position of the vehicle. Then, the lateral offset is calculated
depending on these parameters. In [22], the method proposed
by the authors is limited to static obstacle avoidance. The
execution time which is an important criterion is also not
evaluated. This method can be improved by adapting it in real
Fig. 1. Architecture of the control of the vehicle proposed by [1]. time and relating the planned trajectory distance to the sensors
detection capabilities.
In this study, we focus on the trajectory generation module
by supposing that the ego vehicle, i.e. the car which has B. Horizon planning approach
to perform the lane change, is accurately localized and that The Horizon Planning approach (HP) consists in computing
the perception module provides the distance between the ego the path by dividing the drivable space into convex regions
[15]. The trajectory of each region is computed as the vehicle the next navigation area is automatically performed as the
moves forward. The main advantage is the low execution vehicle is moving forward. Thanks to the perception layer, the
time and the reactivity to the obstacle movement. This allows obstacle position is defined at every step. When considering
quickly computing trajectories and divides the problem into moving objects, the next sampling trajectory is recalculated to
sub-problems but requires treating the continuity problem be adapted to the new obstacle position.
between each planned path. HP approach is widely used in
B. Rolling horizon planning approach
the case of autonomous unmanned vehicles [22]. This method
is useful when the environment is partially unknown. The The horizon length depends on the detection range of the
dimension of each horizon must be correlated to the sensors sensors and can be configured. It can be also correlated to
perception range in order to be reactive to the sensed obstacles the speed of the car and be adaptive at real time. All horizon
or the leading vehicle to be avoided. vectors have the same length (Fig. 4); at the first iteration
(iteration 1), the 1st horizon vector which represents the
III. PROPOSED APPROACH next local planned trajectory is calculated. When the vehicle
position reaches the horizon step distance, the second horizon
A. Principle of the approach
path is calculated for the next planned trajectory. Notice that
The approach developed in this paper extends the use of a common part of each vector is recalculated at each iteration
the sigmoid-based path planning either to static or dynamic in order to update the trajectory for the obstacles position
obstacles perceived in the driving scene. Our proposed algo- changes.
rithm is based on the two previously mentioned principles by
using a configurable sigmoid function to provide an efficient
and feasible path to the vehicle in a given horizon. The path
takes into consideration the position and the motion of the
perceived obstacles or the leading vehicle.
The algorithm is constituted of different parts able to react
to any detected obstacle on the road. The initial reference
path of the vehicle can be provided from a digital map
database or a recorded map. The inputs of the algorithm are
the vehicle position and the obstacle positions which represent
the perception and localization parts as illustrated in Fig. 3. Fig. 4. The proposed rolling horizon planning approach.
The local path is then calculated depending on the desired
The horizon step distance must be also well chosen as
parameter of smoothness and the two parameters of safety
it concerns the distance from which the trajectory must be
distance to be respected during the avoidance manoeuvre. This
recalculated and updated. The ratio between the horizon length
parametric smooth variation is necessary to maintain a level
and the step length represents the number of points to be
of safety and comfort to the passenger of the vehicle and to
calculated in each horizon vector. Evidently, when the step
respect the vehicle dynamics when doing a manoeuvre. The
length is small, the position vector (X, Y ) at each horizon
safety distance when avoiding obstacle can be configurable by
contains more points and will be more precise in the generation
taking into consideration the vehicle velocity. The way-point
of the trajectory. The algorithm finishes when the last horizon
calculated represents the position to reach when avoiding the
vector reaches the limit of the complete trajectory vector. In
obstacle/vehicle based on the generated avoidance path.
the next sub-section, the steps of the algorithm are detailed.
C. Execution steps
The algorithm is decomposed into different steps as illus-
trated in Fig. 5.
The first step is to sense the existence of obstacles in
the scene. If an obstacle is detected, its position (X0 , Y0 )is
recorded. Then, the distance separating the initial local path
from the first obstacle is calculated. And for each horizon, this
vector of the Euclidean distances d(N ) between the vector
of the initial trajectory (X(N ), Y (N )) and the obstacle, is
calculated (N represents the index of the elements of the
horizon vector). The vector of the circular trajectory Yr (N )
is then generated by applying the equality between d(N ) and
Fig. 3. Path-planning strategy.
the desired lateral safety distance S if d(N ) ≤ S (Fig. 6).
The goal is to build a safety circular area around the obstacle
The main objective is to calculate the next local path via
following (2):
the Rolling Horizon Planning Approach depending on the p
perception area of the vehicle sensors. The calculation of Yr (N ) = (S 2 − (X(N ) − X0 )2 ) + Y0 . (2)
 IV. SIMULATION RESULTS
In this section, the simulation results of the obstacle
avoidance method for left-hand overtaking direction will be
presented in different scenarios. The right-hand overtaking
direction is also possible by modifying the way-point of the
curve to the right side of the obstacle. The execution time
which represents a crucial parameter when implementing the
approach in real time is evaluated. The simulations presented
in this section are carried out using Matlab software. The
computer used in simulation has a processor Intel® Core™
i7 − 7820HQ CPU @ 2.90 GHz and 16 Go of RAM.
A. Influence of the smoothness parameter on the curvature
The parameter C in equation (3) has a direct influence on
Fig. 5. Execution steps of the algorithm.
the smoothness of the generated path. This will provide an
adaptive curvature of the path which can be correlated to the
vehicle speed and dynamic behaviour to provide a feasible path
for the vehicle. Simulation tests have been conducted using
two values of parameter C = 0.5 and C = 1. The smaller
value of C gives a smoother curve as shown in the Fig. 7.
This simulation illustrates also the influence of this parameter
on the curvature of the trajectory. The maximum curvature ρ
is reduced from 0.2 for C = 1 to ρ = 0.06 for C = 0.5.
As depicted in Fig. 7, the curve reaches the point P2 in Sm
meters before the obstacle which is chosen at 10 meters for
Fig. 6. Illustration of the initial and the avoiding trajectories in a local path this example. The safety lateral deviation S is chosen to be 4
and the way-point determination in presence of an obstacle. meters.

The next step is the generation of a parameterized smooth

trajectory linking the current vehicle position (P1) to the target
𝑺 P2
way-point position (P3). This point guarantees an imposed 𝑺𝒎
safety distance with a smooth overtaking manoeuvre. The m
obstacle avoidance path Ym (N ) (3) is calculated to reach P3:
Yw
Ym (N ) = (C(d(N )−Sm )))
(3)
(1 + e
where C is a smoothness coefficient (C ∈ [0 1]) of the
exponential function, Sm is the longitudinal safety distance
determining the avoidance manoeuvre distance and d(N ) is
the distance between the initial trajectory and the obstacle.
Fig. 7. Influence of the smoothness parameter C on the path avoidance
D. Path evaluation trajectory and the curvature, two values of C are used for the simulation,
C = 1 and C = 0.5.
The possibility to implement the approach in the vehicle
depends mainly on the respect of the lateral vehicle dynamics
limitation. One of the important parameters of the trajectory is B. Avoidance of a dynamic object (overtaking)
the curvature ρ, it must be continuous and upper bounded. This The simulation of the overtaking manoeuver is investigated
parameter is expressed by the equation (4) which represents in this sub-section, the ego vehicle is placed at the position
the ratio between the angle of the tangent to the path θ and (X, Y ) = (0, 5) meters at the beginning of the simulation
the distance between two consecutive coordinates as follows: whereas the leading vehicle to avoid is placed at (X0 , Y0 ) =
∆θ (30, 5) meters of the ego vehicle. While the whole trajectory
ρ= p [m−1 ]. (4) considered for simulation is 60 meters, the rolling horizon is
(∆X)2 + (∆Y )2
set at 20 meters with the horizon step of 0.5 meter. So, every
The angle of the tangent to the path is calculated using the 0.5 meters, a new trajectory vector Ym (N ) is generated. The
following expression: calculation of the path takes into account several parameters:
the obstacle coordinates, the smoothness parameters of the
 
−1 ∆Y
θ = tan . (5) sigmoid function C, the lateral and the longitudinal safety
∆X
distance. The initial trajectory is considered as a straight line. calculation of each calculated horizon containing (20/0.5) =
When executing the generation of the new path, the position 40 points corresponding to 20 meters of the planned trajectory.
of the dynamic obstacle is assigned at (X0 , Y0 ) = (30, 5)
meters and moves to achieve X = 37.5 meters at the end of
the simulation. The track of the leading vehicle is presented
in grey color. The smoothness parameter C is imposed to 1
and the lateral safety distance S is equal to 4 meters and
the longitudinal safety distance Sm is equal to 10 meters. The
chosen rolling horizon is 20 meters but it can be easily adapted
according to the sensor detection range.
Fig. 8 illustrates the simulation results of the previously Fig. 9. Execution time of each horizon step to calculate the trajectory
mentioned situation. The three consecutive simulations show presented in Fig. 8. The average of the execution time is 43 µs to calculate
three situations: before, during and after the overtaking ma- 20 meters of trajectory.
noeuvre. At the beginning of the simulation (Fig. 8.a), the
leading vehicle is not yet detected as the rolling horizon is 20 C. Real map simulation
meters (green line). Subsequently, when the leading vehicle is
A real trajectory has been recorded at the Museum Cité de
detected, the new trajectory is generated and adapted in real
l’Automobile in Mulhouse to test the developed approach as
time. Even if the obstacle is moving, its position is sensed and
illustrated in Fig. 10. An obstacle is placed on the map with
the way-point is updated by tracking it. The generated path is
given coordinates (X0 , Y0 ). From the obstacle position and the
evaluated by calculating the curvature which is smooth and
reference trajectory, we applied our developed method which
does not reach 0.2 m−1 .
consists in avoiding the obstacle by guarantying imposed
safety distances. The parameters of the sigmoid function used
for the simulation test are: C = 0.2, lateral safety distance
S = 3 m and the longitudinal safety Sm = 15 m.

(a)

(b)

(c)
Fig. 10. Simulation results of a real path by placing a circular shaped obstacle
on the reference trajectory plotted in blue (a). The new path after applying
the avoiding method is in red. In (b) a zoom-in of the avoidance part of the
path. A blue grid is plotted on the figure with a resolution of 5 meters.

In order to validate the proposed avoidance path with respect

(d)
to the vehicle dynamics, the trajectory presented in Fig. 10 is
implemented on a virtual car in CarMaker software (IPG Au-
Fig. 8. Simulation results of the overtaking manoeuvre, the position of the tomotive). The avoidance trajectory is tested with a guidance
leading vehicle is varying. During the overtaking manoeuver, three simulations
are consequently showed before (a), during (b) and after (c) the manoeuvre. controller developed in the laboratory based on robust state
The curvature of the generated path is plotted on (d). feedback coupled with feed-forward action [23]. This test is
performed at 40 km/h (the maximal speed guaranteeing the
The execution time of each path vector (Fig. 9) is computed vehicle stability for this test road) and the results are plotted
at each new step, the number of calculated horizons is (60 − in Fig. 11. The figure time axis is limited to the zoomed
20)/0.5 = 80 calculated values for 20 meters of horizon and trajectory in Fig. 10 in order to show vehicle dynamics around
0.5 m of step length. After five simulations, the average of the the obstacle avoidance. We notice that the curvature followed
execution time of each horizon is very low and equal to 43 by the vehicle measured by the CarMaker follows the road
µs with 2 µs of standard deviation. This time represents the trajectory. The lateral acceleration of the car changes slightly
and is limited between 0 and -2 ms−2 . It is noted also that [4] J. Larson, M. Bruch, R. Halterman, J. Rogers, and R. Webster, “Ad-
the variation of the yaw rate and the steering angle is low and vances in autonomous obstacle avoidance for unmanned surface vehi-
cles,” Space and Naval Warfare Systems Center, San Diego, CA, Tech.
the curves are continuous which enhance passenger comfort. Rep., 2007.
[5] J. Larson, M. Bruch, and J. Ebken, “Autonomous navigation and obstacle
avoidance for unmanned surface vehicles,” in SPIE Unmanned Systems
Technology VIII, Defense Security Symposium, Orlando, FL, USA, 2006.
[6] H. Xu, L. Gao, J. Liu, Y. Wang, and H. Zhao, “Experiments with
obstacle and terrain avoidance of autonomous underwater vehicle,” in
the OCEANS 15 MTS/IEEE, Washington, DC, USA, 2015, pp. 1–4.
[7] S. Skiena, “Dijkstras algorithm,” Implementing Discrete Mathematics:
Combinatorics and Graph Theory with Mathematica, Reading, MA:
Addison-Wesley, pp. 225–227, 1990.
[8] A. Bry and N. Roy, “Rapidly-exploring random belief trees for motion
planning under uncertainty,” in the IEEE International Conference on
Robotics and Automation (ICRA), Shanghai, China, 2011, pp. 723–730.
[9] F. Von Hundelshausen, M. Himmelsbach, F. Hecker, A. Mueller, and
H. Wuensche, “Driving with tentacles: Integral structures for sensing
and motion,” Journal of Field Robotics, vol. 25, no. 9, pp. 640–673,
Fig. 11. Vehicle dynamics and curvatures plotted as a function of time and 2008.
limited around the avoidance manoeuver [10] H. Mouhagir, V. Cherfaoui, R. Talj, F. Aioun, and F. Guillemard,
“Using Evidential Occupancy Grid for Vehicle Trajectory Planning
Under Uncertainty with Tentacles,” in the 20th IEEE International
V. CONCLUSION AND FUTURE WORK Conference on Intelligent Transportation, Yokohama, Japan, 2017, pp.
In this paper, the light was shed on path planning of 1–7.
[11] S. Glaser, B. Vanholme, S. Mammar, D. Gruyer, and L. Nouveliere,
autonomous vehicles. A new method of obstacle avoidance “Maneuver-based trajectory planning for highly autonomous vehicles
and overtaking has been presented. The idea consists in on real road with traffic and driver interaction,” IEEE Transactions on
proposing a safe path for the vehicle instead of the initial Intelligent Transportation Systems, vol. 11, no. 3, pp. 589–606, 2010.
[12] D. Gonzalez, J. Perez, R. Lattarulo, V. Milanes, and F. Nashashibi,
trajectory. The method is based on the use of a parametric “Continuous curvature planning with obstacle avoidance capabilities
sigmoid function coupled with a rolling horizon approach in urban scenarios,” in The 17th International IEEE Conference on
to divide the trajectory in convex regions. Each horizon is Intelligent Transportation Systems (ITSC), Qingdao, China, 2014, pp.
1430–1435.
calculated and updated at each step.This is then used to [13] J. Daniel, A. Birouche, J.-P. Lauffenburger, and M. Basset, “Navigation-
calculate online a portion of the path. The most interesting based constrained trajectory generation for advanced driver assistance
features of this method are the configurable smoothness of the systems,” International Journal of Vehicle Autonomous Systems, vol. 9,
no. 3-4, pp. 269–296, 2011.
trajectory and the very low execution time. The lateral and the [14] M. Arbitmann, U. Sthlin, M. Schorn, and R. Isermann, “Method and
longitudinal safety distances between the ego vehicle and the device for performing a collision avoidance maneuver,” US Patent
avoided obstacle can be easily parameterized depending on US8 209 090B2, 2012.
[15] X. Wang and H. Kopfer, “Dynamic collaborative transportation planning:
the speed of the obstacle. The developed approach has been A rolling horizon planning approach,” in the 4th International Confer-
validated by simulating different situations. The results of lane ence on Computational Logistics (ICCL’13). Copenhagen, Denmark:
change when avoiding static or dynamic obstacles on a real Springer, 2013, pp. 128–142.
[16] T. Schouwenaars, r. Fron, and J. How, “Safe receding horizon path
recorded map prove the validity of our strategy. We plan to planning for autonomous vehicles,” in the 40th Annual Allerton Confer-
extend our algorithm for multi-obstacle avoidance as it is a ence on Communication, Control, and Computing, Monticello, IL, USA,
common situation for overtaking manoeuvres. We are actually 2002, pp. 295–304.
[17] J. Ziegler, P. Bender, T. Dang, and C. Stiller, “Trajectory planning for
working on the correlation of the parameters of the sigmoid BerthaA local, continuous method,” in the IEEE Intelligent Vehicles
function to the vehicle speed and to the maximum allowed Symposium (IV), Michigan, USA, 2014, pp. 450–457.
curvature and their real time adaptation. Further research will [18] A. Broggi, P. Cerri, S. Debattisti, M. C. Laghi, P. Medici, M. Panciroli,
and A. Prioletti, “Proud-public road urban driverless test: Architecture
be focused on the integration of the path modification with the and results,” in the IEEE Intelligent Vehicles Symposium (IV), Michigan,
lateral control of the vehicle and their implementation on our USA, 2014, pp. 648–654.
laboratory autonomous car ARTEMIPS. Finally, considering [19] T. Gu, J. M. Dolan, and J.-W. Lee, “On-road trajectory planning for
general autonomous driving with enhanced tunability,” in Intelligent
this configurable low parameters method, a comparison with Autonomous Systems. Springer, 2016, vol. 13, pp. 247–261.
state-of-the-art local planning solutions will also be performed. [20] A. Y. Hata, F. S. Osorio, and D. F. Wolf, “Robust curb detection and
vehicle localization in urban environments,” in the IEEE Intelligent
R EFERENCES Vehicles Symposium (IV), Michigan, USA, 2014, pp. 1257–1262.
[1] R. Attia, J. Daniel, J.-P. Lauffenburger, R. Orjuela, and M. Basset, “Ref- [21] A. Hata and D. Wolf, “Road marking detection using LIDAR reflective
erence generation and control strategy for automated vehicle guidance,” intensity data and its application to vehicle localization,” in the 17th
in the IEEE Intelligent Vehicles Symposium (IV), Madrid, Spain, 2012, International IEEE Conference on Intelligent Transportation Systems,
pp. 389–394. Qingdao, China, 2014, pp. 584–589.
[2] O. Esrafilian and H. D. Taghirad, “Autonomous flight and obstacle [22] S. Vera, J. Cobano, G. Heredia, and A. Ollero, “Rolling-horizon trajec-
avoidance of a quadrotor by monocular SLAM,” in the 4th International tory planning for multiple UAVs based on pseudospectral collocation,”
Conference on Robotics and Mechatronics (ICRoM), Tehran, Iran, 2016, in the International Conference on Unmanned Aircraft Systems (ICUAS),
pp. 240–245. Orlando, FL, USA, 2014, pp. 516–523.
[3] M. Fraiwan, A. Alsaleem, H. Abandeh, and O. Aljarrah, “Obstacle [23] M. Boudali, R. Orjuela, and M. Basset, “A Comparison of Two Guidance
avoidance and navigation in robotic systems: A land and aerial robots Strategies for Autonomous Vehicles,” in the IFAC 2017 World Congress,
study,” in the 5th International Conference on Information and Commu- Toulouse, France, 2017, pp. 12 539–12 544.
nication Systems (ICICS), Irbid, Jordan, 2014, pp. 1–5.