Proceedings of the 2018 IEEE

International Conference on Robotics and Biomimetics

December 12-15, 2018, Kuala Lumpur, Malaysia

Trajectory Tracking of Unicycle-type Robots with Constraints

Chaoxun Guo1, Zhenglong Sun2, Yongquan Chen2, Yutong Xie2, Shuaijun Li1, Huihuan Qian2

Abstract—Trajectory tracking is a decisive factor that affects present, geometry based approach is one of the main methods
autonomous driving performance of mobile robots in real-world to solve this kind problem. Map is divided into grids, and an
applications, such as industry, service and other areas. Nowadays, optimal path to reach the target point is searched in grids.
there has been many works about path planning and trajectory Furthermore, lots of path planning algorithms have been
tracking. Whereas there are problems to be improved to make proposed such as Dijkstra, A* [5] [6]. Dijkstra algorithm fixes
mobile robot well applied in practical world. For example, a single node as the "source" node and finds the shortest path
generated trajectory can’t be tracked smoothly, and the robot from the source to all other nodes in the graph, producing a
has to stop and rotate itself to adjust pose and pass segments in shortest-path tree. In contrast, Best First Search (BFS) is better
trajectory because of constrains of path curvature and robot’s
than Dijkstra because it refers to a search with a heuristic that
kinematics model. In this paper, a new trajectory planning
algorithm with cubic spline interpolation method was proposed,
attempts to predict how close the end of a path is to a solution.
which fully considered constrains of robot model including However, A* algorithm adopts both the ascendancies of
velocity and acceleration bounds. And corresponding trajectory Dijkstra and BFS, and it is widely used. Moreover, A*
was generated. Then a robust controller was built to conduct algorithm can generate an optimal and efficient path between
tracking tests. Experimental results demonstrate that our the starting and the ending points. Besides, some other
planning and tracking algorithm could help robot achieve a algorithms were developed and became focus, such as Depth
smooth, accurate and efficient trajectory tracking performance First Search (DFS), Probalbilistic Roadmao Method (PRM)
without any stop. [7], Rapidly-Exploring Random Tree (RRT) [8]. Stanford put
Keywords—Path planning, Trajectory generation; Trajectory forward a mixed A algorithm which considers
Tracking; Unicycle-type Mobile Robot; Controller; four-dimensional space and the orientation of robots.
It was noted that path generated with global planning
I. INTRODUCTION
algorithm or search-based is discrete, it means that there may
Mobile robot is increasingly demanded in industry, be two or more sharp turns or zigzag which are not tractable
delivering, service and other areas. A good trajectory tracking for unicycle robots. Once robot moves along a path with
performance is desired in practical applications for mobile specified points, curvature or limitations, it has to stop and
robots to work effectively and move along an expected route. rotate itself to adjust pose position and orientation to pass
Normally, this process includes path generation and trajectory specified segment and continue its journal. In the past decades,
tracking control, which are highly related to planning researcher studied path planning from following aspects such
algorithm and trajectory tracking controller. Recently, works as smoothness, low cost consumption, efficiency, stability, run
have been done on these two fields, and many algorithms and time and path length [9] [10]. Indeed, smooth path was
controller were designed. There are still many researchers investigated. Approaches to obtain smooth path were
focus on these areas [1], [2], [3]. Besides, the geometric model proposed in terms of geometry methods. Researches about
of robot was taken into consideration to improve tracking polynomial curve, clothoids, Bezier curves were conducted
performance [4]. However, reported works does not address [11]. In [12], authors addressed a planning algorithm with
the constraints of kinematics and dynamics model of robot so optimal and smoothness effect. In these smoothness methods,
it’s still challenging to achieve an accurate and smooth few of them paid enough attention to the effect of velocity and
trajectory tracking performance. acceleration bounds to gain a good tracking ability. Some of
them only simply optimize the path to be a smooth one without
Path planning is to search a best path from the starting taking constraints of robot into account. A smooth path can’t
point to the end without any collision with obstacles. At be regarded as one that can be well tracked because of
dynamics and kinemics limitations of mobile robot.
*This work was supported by Economic, Trade and Information
Commission of Shenzhen Municipality (Grant No. ZLZBCXLJZI201608050 Trajectory tracking control is an essential link of the
20016), and also financially supported by Innovation and Technology autonomous mobile robot. Controller adjusts angular velocity
Commission of Hong Kong (Grant No. I73/0/1AP), Shenzhen Science
and linear velocity in time dimension, which is effect to
and Technology Innovation Commission (Grant No. JCYJ2017081810450
2599), and partially supported by the National Natural Science Foundation of eliminate the error between the reference trajectory and
China (Grant No: U1613216) tracking trajectory in real world. Nowadays, many researchers
1
Chaoxun Guo, Shuaijun Li are with The Chinese University of Hong focus on tracking control problem. Existing control methods
Kong, Shatin, N.T,Hong Kong SAR, China (e-mail: chaoxunguo@cuhk. include adaptive control [13], Backstepping Control, and
edu.hk, [email protected]). Sliding mode control [14], etc. In [15], Lyapunov controller
2
Zhenglong Sun, Yongquan Chen, Huihuan Qian, Yutong Xie are with
setting method should be adopted to get a stable result. In [16],
The Chinese University of Hong Kong Shenzhen, Shenzhen, China a necessary condition of Brockett smooth state feedback
(email:{sunzhenglong;yqchen;hhqian}@cuhk.edu.cn,[email protected] stabilization is the theoretical support of motion control of
k.edu.cn ).

978-1-7281-0377-8/18/$31.00 © 2018 IEEE 1700

mobile robots. In [17], a global time-varying trajectory velocity is in range [vmin, vmax], angular velocity is in range
tracking control law was designed, which is an early global [Zmin, Zmax]. Speed limitation is a constraint for unicycle-type
result. Basing on Adaptive Backstepping, authors of [18] paid robot to move smoothly along the trajectory. Usually a path
attention to the adaptive control of mobile robots under without any stop in the tracking process is defined as a good
uncertainties. In [19], the author studied the non-holonomic trajectory.
dynamic system after chain transformation. Similarly, in [20],
authors designed a tracking controller with uncertain inertial
parameter in the dynamic system. Also, researchers [21]
combined Backstepping with the time-varying feedback to
propose a robust tracking control law of exponential
convergence respectively. Most tracking control strategies
focused on how to improve the tracking accuracy and
efficiency, and the study was based on an assumed trajectory.
Whereas little attention was paid on how the reference
trajectory was produced.
During past decades, works has been done about path
planning and trajectory tracking. While problems caused by
constraints or motion limitation was not fully addressed.
Aiming to improve smoothness of the trajectory and obtain an Fig. 1 Trajectory Tracking of a Unicycle-type Robot
efficient tracking performance, we did this research. In this
paper, we chose unicycle robot, a kind of mobile robot, as our B. Path Improvement with Constrains
research object. A new path planning algorithm was designed, Trajectory generated with A* or other algorithms may not
which ensures that mobile robot moves smoothly without be well tracked because constraints mentioned above.
constraints from capability of the robot itself. Additionally, an Ignorance of unicycle model also weakens the tracking
accurate and efficient controller was adopted. Experiments performance. A generated path shall be improved to become a
showed that the combination of this controller and path permanent tractable trajectory. In spatial dimension, a path
planning algorithms can achieve a smooth and efficient consists of a set of {Pi=(xi,yi)}, representing only spatial
trajectory performance without termination or stop, and the information; while (vi,Zi) represents robot’s linear velocity and
trajectory in real world well overlapped the reference. angular velocity at a spatial point. In time dimension, a
The rest of the paper is organized as follows. In Section II, trajectory can be viewed as a set of {Pti=(xti,yti,Tti)}
the characteristics of nonholonomic unicycle model were representing robot’s position and orientation in time
discussed. And constraints to path generation and trajectory dimension; (vti,Zti) represents robot’s linear velocity and
tracking were analyzed. Then a new path planning method angular velocity at a specific time moment.
taking all constrains into account was proposed, and the Assume Z ę [Zmin,Zmax], v ę [vmin, vmax], Trajectory
enhanced tracking performance was fully explained. In tracking can be simplified as rotation movement in many small
Section III, tracking controller was built according to the segments, in which robot rotates around center C. Denoting the
moving speed of the mobile robot, and the controller is
radius as U and curvature of trajectory as N, then the maximum
verified by tests. Section IV, trajectory tracking performance
of curvature can be calculated with formulation (1), and Nę
with a new path planning algorithm and designed controller
was analyzed. In Section V, conclusions were drawn and [0,Nmax]. Then it can be concluded that geometrical curvature
future direction was discussed. in bounded scope and continuous velocity are two constraints
for an ideal tractable trajectory
II. TRAJECTORY GENERATION AND IMPROVED TRACTABILITY ­ vti
° Zti N .vti
A. Constraints of Unicycle-type Robot ® U ti ˄1˅
°
¯N max Zmax / vmin
Unicycle-type robot is a kind of mobile robots which is
widely used in our life. So we chose it as research object in this
paper. By assigning velocities to two wheels, a specified linear Basing on a path generated with global path planning
velocity and angular velocity are produced, which means that algorithm, we proposed a global path planning method. As
robot can’t move along lateral direction. shown in Fig 2, the generated path contains many sharp turns,
so we have to make some improvement on it. The first step is
Two main constraints result in the generated trajectory is not
to smooth the sharp points with cubic spline interpolations,
able to be well tracked. One is that the path contains points
which models the path using piecewise third order
needing lateral movement. For example, in Fig 1, a robot is
polynomials, then the original path becomes smooth without
moving along a reference trajectory. Without a lateral velocity,
sharp corners. The path can be denotes as
robot has to stop to rotate itself while arriving at zigzag point A.
y=fi(x)=aix3+bix2+cxi+di, the modified path is the blue curve in
In continuous and smooth curve OA, there might be also some
Fig 2. In addition, a continuous v is obtained:
points requiring a lather velocity. Another factor is the
fi’(x)=3aix2+2bix+ci.
physical constraint from robot model. For mobile robot, linear

1701
 are assigned an appropriate angular velocity and linear
velocity.
Another main step of trajectory generation is to transform
spatial information into time dimension. Each small segment
in path can be regarded as a small time-interval, and we can
assign velocity in time dimension. For example, PiPi+1 in Fig 1
matches with vi and wi in spatial, the corresponding time
moments are assigned to (vti ,Zti), 't can be calculated by
formulation (2) and (3). We would get a series of time intervals
{'ti}to all the points {Pi} on path, and the starting point.

'ti

³ v cos T  Z t
0
i i i dt xi1  xi Δxi (2)
Fig. 2 Path Adjusting Using Cubic Spline Interpolation

Autonomous navigation robots sometimes are expected to arcsin 'xi ˜ N i  sinT i  T i (3)
Δ ti
pass specified locations to complete tasks, for example, the red Zi
dots in Fig 2. But above improvement methods may miss these
important points. So it is necessary to modify the trajectory
again. We can pick a new set of control points including both III. CONTROLLER DESIGN AND EXPERIMENTS
these specified points and some other points from the old Controller is the core of mobile robot to obtain a good
smoothed path. For example, the task must be done through tracking performance. A controller is chosen for unicycle-type
three red dots, so that the improved curve needs to be robot to track the trajectory. Then, experiments are conducted
reprocessed again. Points in two and three times’ length of red to validate the tracking performance in real world.
dots are selected to be optimized, and the cubic spline
interpolation method is carried out again. The regenerated A. Trajectory Tracking Controller Design
smooth curve is shown as the smooth curve in Fig 2. Finally, Unicycle-type mobile robot, such as AGVs or Turtlbot,
the curvatures of whole path are checked to ensure that all of needs to adjust linear velocity to perform a good tracking
them are in the tractable bound, while abnormal curvature
performance. However, there are limits of vę[vmin, vmax] and
point would be optimized again.
Zę[Zmin, Zmax]. We should take all these constraints into
account when designed unicycle-type robot trajectory tracking
C. Trajectory Generation with Constraints
controller. According to [20], a reference trajectory, [xr(t) yr(t)
According to the improved smooth path, we designed Tr(t) ]T, could be generated by the reference linear velocity and
motion controller based on robot’s capability. One main step is angular velocity [vr(t) Zr(t)]T as in (4),
to assign velocity to generated path. The path was cut into a
great amount of small segments, such as PiPi+1 in Fig 1. Then
the curvature of segments would be calculated with ­ x r =ν r cos θ,
formulation: N=|f’’(x)|/(1+f’(x))3/2. Moreover, vi and Zi can be °
obtained with formulation (1).
® y r =ν r sin θ, (4)
° θ =ω
¯ r r

The tracking error can be measured as a feedback in

position domain as in formulation (5), and the control law can
be designed as formulation (6) to transform the feedback into
velocity domain as control input.

§ xe · § cos T sin T 0 · § xr  x ·
¨ ¸ ¨ ¸¨ ¸
¨ sin T cos T 0 ¸ ¨ yr  y ¸ (5)
¨ ye ¸
¨T ¸ ¨ ¸¨ T  T ¸
© e¹ © 0 0 1 ¹© r ¹
­ c1x e
° ν vr 
° 1  x 2e  y 2e
° (6)
® § θ θ ·
° c2 v r ¨ ye cos e  x esin e ¸
°ω ω r  © 2 2 ¹  c sin θ e
3
Fig. 3 Logical Scheme to Determine Linear and Angular Velocity °¯ 1  x e  ye
2 2 2
In detail, both vi and Zi are determined spatially according This controller for unicycle-type mobile robot is superior
to Fig 3. With logical scheme in Fig 3, all segments in the path
to others because the control inputs will be bounded, vę[vmin,

1702
vmax] and Zę[Zmin, Zmax]; and the control is of fast response
with no time-lagging item. C1, C2 and C3 are positive constants
which can be tuned subject to the velocity constraints. It has
been shown in [22] that this controller can track a variety of
different trajectories accurately.

B. Experiments with Tracking Controller

To validate the controller’s tracking performance, we
conducted tracking experiments with TurtleBot, as shown in
Fig 5. TurtleBot is a kind of unicycle-type robot based on
Robot Operation System (ROS), and its chassis evolves from
Koubuki platform. It consists of two driving wheels. In
addition, TurtleBot has 2D-Lidar sensors used for navigation
and localization in trajectory tracking. 2D-Lidar is used to
obtain real-time location; while the odometer collects the
angular velocity and linear velocity; three-axis digital
gyroscope gets the pose. TurtleBot will return the position in
real world to controller while tracking the trajectory. Then
controller updates and transmits velocity to TurtleBot through
laptop.

Fig. 5 Trajectory Tracking Result with Designed Controller

It can be seen that the deviation is distinctive in the right

side the trajectory. Coefficients as C1, C2 and C3 concern the
tracking performance, both of should be adjusted to win a
better tracking performance. We made several experiments
only to find that both C1 and C3 are more sensitive to the
tracking performance. A very small change in C1, such as 0.3,
can eliminate or cause the large deviation. C2 is suitable for
Fig. 4 TurtleBot Used in Experiments slight adjustment of trajectory. In addition, Different matches
may bring the same result, for example, {C1=1.5, C2=12.5,
To validate the performance of this controller, we would C3=15} has almost the same tracking performance with {C1=3,
compare the real trajectory with the reference trajectory. C2=14.5, C3=5}.
Firstly, we set an origin in ground and built a virtual coordinate
system. Then the TurtleBot could locate itself in real world Through several attempts, we finalized C1=2, C2=14.5,
with 2D-Lindar. Then we designed a reference trajectory in C3=4. Trajectory tracking test was conducted with free starting
Matlab and assigned velocity in time dimension. The reference point, Pi (0.5,0.5) the result was shown in Fig 6. It can be seen
trajectory shares the same virtual coordinate system built in the real route is basically consistent with the reference
real world. trajectory while it was 0.135m away from the start point.
TurtleBot merely moved along the trajectory, which means
In this experiment, a knot-shaped trajectory as in [22] was that the controller could quickly adjust the motion parameters
chosen. The reference trajectory was denoted in velocity space and had a fast response.
as vt=0.4, and a trajectory is generated. Then we conducted
tests on a flat ground with TurtleBot connecting with a laptop. Additionally, the experiment tracked continuously 3 times
The velocity limitation of the robot is constrained, [vmin, along the reference trajectory. Real trajectory indicated a good
vmax]=[0.1,1.5m/s] and Zmax =5rad/s. Values, such as Xe, Ye, x, convergence, and coincided with the origin at the center. It can
y, can be collected from TurtleBot and are transmitted to a be concluded that the controller has strong robustness and
laptop. With all the information from robot, controller updates stability, it can resist interference.
Zti and vti and transmits them to TurtleBot by laptop to the
trajectory. At first, setting the coefficients as C1=5, C2=12,
C3=3.3, tracking result is presented in Fig 5.

1703
 Y(m)
`Fig 7 Path Smoothing using Smoothing Spline Approach
Then we assigned velocity to the path according to the
method mentioned in section II. N in PiPi+1 can be calculated,
and (vi,Zi) can be got. (vi,Zi) was transformed to (vti,Zti) with
formulation (2) and (3). The new trajectory is the blue line
shown in the Fig 8, the original path is red line. We can see that
the trajectory and the path almost overlapped totally. So the
method to generate tractable trajectory works well.

Fig. 6 Trajectory Tracking Performance with Controller

A slight deviation occurred at lateral sides. Xe was

bounded in (-0.133, 0.134), the maximum occurred where
Pi=(-0.48, 3.64); Ye is in scope (-0.132, 0.134), the maximum
occurred where Pi = (0.32, 4.25). However, a slight deviation
didn’t weaken the tracking performance since trajectory
overlapped the reference trajectory tightly. Deviation may
result from two aspects. The first one is that the controller's
correction and adjustment process are discrete and can only be
performed with time interval, which may cause a time-lagging.
Furthermore, 0.05s was needed to update the velocity which
means the velocity was not adjusted timely. Thus deviation
was accumulated. Besides, noise from TurtleBot or slipping on
the ground brings certain errors to real velocity. Although Fig. 8 Trajectory Generation by Assigning Time to Velocities
deviation exists, it is within an acceptable scope, and a good TurtleBot tracked the trajectory with a free start point Pi=
tracking performance is still achieved. (-0.05,-0.08), and the experimental result is shown in Fig 9.
The red curve is a reference trajectory, and the blue one is the
IV. TRAJECTORY TRACKING CONTROL WITH A NEW real trajectory of TurtleBot. After movement of 0.18m in X
PLANNING ALGORITHM direction, TurtleBot quickly closed to reference trajectory. In
In order to further validate the tracking performance of the other words, the controller is almost with no time-lagging. An
new generation algorithm and controller, we conducted adjustment of orientation and velocity made actual trajectory
another experiment. Firstly, we obtained a path generated with coincide with reference again when a deviation occured. In the
global path planning algorithms, as the blue curve shown in whole tracking process, the maximum Xe is 0.2747m and is in
Fig 7. We can see that there are many turns and sharp points. the end. The curve in the end is steep, which means larger
With cubic spline interpolation, we made some improvement acceleration is needed for TurtleBot, however TurtleBot could
on it. The improved path was the red line in Fig 7. The scatter not achieve such high acceleration. In the whole tracking
sharp turns in the original path were connected with a smooth process, real trajectory is identical to the reference trajectory,
line, and no sharp point existed anymore. The path can be saying, the controller and the path planning algorithms are
tracked freely because of the continuous velcity and suitable with high stability and effect.
curvature N in each point.

1704
 [8] A. Gasparetto, P. Boscariol, A. Lanzutti, and R. Vidoni, “Path planning
and trajectory planning algorithms: A general overview,” In Motion and
Operation Planning of Robotic Systems: Background and Practical
Approaches, vol. 29, G. Carbone and F. Gomez-Bravo, Eds. Cham,
Switzerland: Springer-Verlag, 2015, pp. 3–27.
[9] F. Kamil, SH. Tang, W. Khaksar, N. Zulkifli and SA Ahmad “A review
on motion planning and obstacle avoidance approaches in dynamic
environment,” Advances in robotics &Automation, vol. 4, no 2, pp. 1–8,
2015.
[10] D. Dolgov, S. Thrun. “Path planning for autonomous vehicles in
unknown semi-structured environments,” The International Journal of
Robotics Research., vol. 29, no. 5, pp. 485-501, 2010.
[11] A Mohammadi, M Rahimi, AAA Suratgar. ”New path planning and
obstacle avoidance algorithm in dynamic environment”. Iranian
Conference on Electrical Engineering. Tehran, Iran, 2014, pp. 1301 –
1306.
[12] Wenjie Dong and Yi Guo, “New trajectory generation methods for
Fig. 9 Tracking Performance on Improved Trajectory nonholonomic mobile robots”, Proceedings of the 2005
International Symposium on Collaborative Technologies and
Systems, pp. 353-358, 2005.
V. CONCLUSION AND FUTURE WORK [13] W. E. Dixon, Q. M. De, D. M. Dawson. “Adaptive Tracking and
Regulation of a Wheeled Mobile Robot with Controller/LTpdate Law
The main conclusions from the research are drawn as Modularity,” Proc. ICRA, 2002, pp. 2620-2625.
follows. A new trajectory planning algorithm considering the [14] S.G. Tzafestas, K.M. Deliparaschos, G.P. Moustris. “Fuzzy logic path
nonholonomic constraints of a unicycle model was proposed tracking control for autonomous non-holonomicmobile robots: design
for unicycle-type robot. It removed the zigzag point and of system on a chip,” Robotics and Autonomous Systems., vol. 58, no. 8,
pp. 1017-1027, 2010.
curvature constrains in the trajectory, making the path become
[15] C. Samson, K. Ait-Abderrahim. Feedback control of a nonholonomic
a smooth and easily-tracked one. With this method, the wheeled cart in cartesian space. Proc. ICRA, 1991, pp. 1136-1141.
trajectory which generated by global path planning algorithm [16] R. W. Brockett. ”Asymptotic stability and feedback stabilization,”.
becomes highly smooth and trackable for mobile robots. In Differential Geometric Control Theory,. vol. 27, pp.181-191, 1983.
addition, a fast and accurate controller for trajectory tracking [17] G. Walsh, D. Tilbury, S. Sastry, et al. “Stabilization of trajectories for
was adopted. By applying the controller to improved trajectory, systems with non-holonomic constraints”. IEEE Transactions on
experiments show that the trajectory tracking with new Automatic Control., vol. 39, no. 1, pp. 216-222, 1994.
algorithm achieves a robust tracking performance. [18] W. Dong. “On trajectory and force tracking control of constrained
mobile manipulatorswith parameter uncertainty,” Automatica., vol. 38,
This paper provides a solution to trajectory tracking of no. 9, pp. 748-760, 2002.
mobile robots, enhanced their tacking ability, which promotes [19] W. Dong, W. Huo, S. K. Tso. “Tracking control of uncertain dynamic
non-holonomic system and its application to wheeled mobile robots,”
the application of robots in real world. In this work, the IEEE Transactions on Robotics and Automation., vol. 16, no. 6, pp.
constants of controller, such as C1, C2, and C3, are validated 870-874, 2000.
through several attempts before used. For future work, we will [20] J. T. Huang. “Adaptive tracking control of high-order non-holonomic
design an intelligent controller which can adjust constants mobile robot systems,” IET Control Theory & Applications, vol. 3, no.
6, pp. 681-690, 2009.
automatically in different segment of the trajectory according [21] Z. P. Jiang, H. Nijmeijer. “Tracking control of mobile robots: a case
to the curvature and other geometry parameters. Thus the robot study in backstep-ping,” Automatica., vol. 33, no. 7, pp. 1393-1399,
can keep an effect performance in the whole tracking process. 1997.
[22] X. Yu, L. Liu, and G. Feng, “Trajectory tracking for nonholonomic
vehicles with velocity constraints,” IFAC-PapersOnLine., vol. 48, no.
REFERENCES 11, pp. 918–923, 2015.
[1] A Stentz. “Optimal and efficient path planning for partially-known
environments,” Proc. ICRA, San Diego, USA, 1994, pp. 3310-3317.
[2] Z.-P. Jiang, “Controlling underactuated mechanical systems: A review
and open problems,” in Advances in the theory of control, signals and
systems with physical modeling. Springer, 2010, pp. 77–88
[3] K. Wang, Y. Liu, and L. Li. “Visual servoing trajectory tracking of
nonholonomic mobile robots without direct position measurement,”
IEEE Trans. Rob., vol. 30, no. 4, pp. 1026–1035, 2014.
[4] J. Snider, “Automatic steering methods for autonomous automobile
path tracking,” Tech. Rep. CMU-RI-TR-09-08, Robotics Institute,
Carnegie Mellon University, Pittsburgh, PA, USA, 2009.
[5] S. J. Russell, P. Norvig. “A Modern Approach,” in Artifiacial Intelligent,
3rd Ed. USA: Prentice Hall, 2009, pp. 97-129.
[6] P E Hart, N J Nilsson, B Raphael. “A formal basis for the heuristic
determination of minimum cost paths.” IEEE Trans of Systems Science
and Cybernetics, vol. 4, no. 2, pp. 100-107, 2007.
[7] P Jensfelt, S Kristensen. “Active global localization for a mobile robot
using multiple hypothesis tracking.” IEEE Transactions on Robotics
and Automation., vol. 17, no. 5, pp. 748-760, 2001.

1705