IEEE TRANSACTIONS ON ROBOTICS, VOL. 23, NO.

3, JUNE 2007 519

Vector Field Path Following for Miniature

Air Vehicles
Derek R. Nelson, D. Blake Barber, Timothy W. McLain, Senior Member, IEEE, and
Randal W. Beard, Senior Member, IEEE

Abstract—In this paper, a method for accurate path following achievable tracking precision. For MAVs, wind speeds are com-
for miniature air vehicles is developed. The method is based on monly 20%–60% of the desired airspeed. Effective path tracking
the notion of vector fields, which are used to generate desired strategies must overcome the effect of this ever present distur-
course inputs to inner-loop attitude control laws. Vector-field
bance. For most fixed-wing MAVs, the minimum turn radius is
path-following control laws are developed for straight-line paths
and circular arcs and orbits. Lyapunov stability arguments are in the range of 10–50 m. This places a fundamental limit on
used to demonstrate asymptotic decay of path-following errors in the spatial frequency of paths that can be tracked. Thus, it is
the presence of constant wind disturbances. Experimental flight important that the path-tracking algorithms utilize the full ca-
tests have demonstrated mean path-following errors on less than pability of the MAV. Finally, high-resolution state sensors with
one wingspan for straight-line and orbit paths and less than three high-frequency updates are not typically available for MAVs.
wingspans for paths with frequent changes in direction. Successful tracking approaches must exploit fully those sensors
Index Terms—Miniature aerial vehicle (MAV), path following, that are readily available.
trajectory tracking, unmanned aerial vehicle (UAV). Several approaches have been proposed for UAV trajectory
tracking. An approach for tight tracking of curved trajectories
is presented in [1]. For straight-line paths, the approach approx-
I. INTRODUCTION
imates PD control. For curved paths, an additional anticipatory
NMANNED aerial vehicles (UAVs), large and small, have
U demonstrated their usefulness in military applications.
Furthermore, there are numerous potential uses for UAVs in
control element that improves the tracking capability is imple-
mented. The approach accommodates the addition of an adap-
tive element to account for disturbances such as wind. This ap-
civil and commercial applications and the prospects for broad proach is validated with flight experiments.
impact are strong. To extend the usefulness of UAVs beyond In [2], Kaminer et al. describe an integrated approach for de-
their current applications, the capability to plan paths and to veloping guidance and control algorithms for autonomous ve-
follow them accurately is of great importance. Unlike piloted hicle trajectory tracking. Their approach builds upon the theory
vehicles, which rely on the pilot to navigate over demanding of gain scheduling and produces controllers for tracking trajec-
terrain or to avoid obstructions, UAVs rely on automation tories that are defined in an inertial reference frame. The ap-
to provide this functionality. As applications such as urban proach is illustrated through simulations of a small UAV.
surveillance and rural search and rescue require UAVs to fly Implicit in the notion of trajectory tracking is that the vehicle
down city streets surrounded by buildings or near the surface is commanded to be in a particular location at a particular time
of abruptly changing mountainous terrain, the ability to follow and that this location typically varies in time, thus causing the
preplanned paths with precision is essential. For missions vehicle to move in the desired fashion. With fixed-wing MAVs,
involving cooperation among a team of UAVs, precise path the desired position is constantly moving (at the desired ground
tracking is often crucial to achieving the cooperation objective. speed). The approach of tracking a moving point can result in
For miniature aerial vehicles (MAVs),1 such as those of pri- significant problems for MAVs if disturbances, such as those
mary interest in this study, wind disturbances, dynamic char- due to wind, are not accounted for properly. If the MAV is flying
acteristics, and the quality of sensing and control all limit the into a strong wind (relative to its commanded ground speed),
the progression of the trajectory point must be slowed accord-
Manuscript received December 14, 2005; revised September 15, 2006. This ingly. Similarly, if the MAV is flying down wind, the speed of
paper was recommended for publication by Associate Editor G. Sukhatme and
Editor K. Lynch upon evaluation of the reviewers’ comments. This work was
the tracking point must be increased to keep the MAV from over-
supported by AFRL/MNK Grant F08630-03-1-0017 and by the Air Force Office running the desired position. Given that wind disturbances vary
of Scientific Research under Grant FA9550-04-1-0209 and Grant FA9550-04-C- and are often not easily predicted, trajectory tracking can be very
0032. challenging in anything other than calm conditions.
D. R. Nelson is with the Unmanned Systems Division, Northrop Grumman
Corporation, San Diego, CA 92127 USA. Rather than pursuing the trajectory tracking approach, this
D. B. Barber and T. W. McLain are with the Mechanical Engineering Depart- paper explores path following where the objective is to be on
ment, Brigham Young University, Provo, UT 84602 USA (e-mail: mclain@byu. the path rather than at a certain point at a particular time. With
edu).
R. W. Beard is with the Electrical and Computer Engineering Department, path following, the time dependence of the problem is removed.
Brigham Young University, Provo, UT 84602 USA. In [3] and [4], performance limits for reference-tracking and
Digital Object Identifier 10.1109/TRO.2007.898976 path-following controllers are investigated and the difference
1We consider MAVs to be those with wingspans in the 0.3–2-m range and
between them is highlighted. It is shown that there is not a fun-
micro aerial vehicles to be those with wingspans under 0.3 m. Here, the abbre- damental performance limitation for path following for systems
viation MAV denotes miniature aerial vehicle. with unstable zero dynamics as there is for reference tracking.
1552-3098/$25.00 © 2007 IEEE
520 IEEE TRANSACTIONS ON ROBOTICS, VOL. 23, NO. 3, JUNE 2007

Fig. 1. Illustration of the vector field idea for straight-line and circular path following. The desired course of the MAV is specified the by direction of the vector field.

Building on the work presented in [5] on maneuver modi- of a potential field to navigate through heavily populated areas
fied trajectory tracking, Encarnação and Pascoal develop an ap- safely while still aggressively approaching their targets. Vector
proach that combines the features of trajectory tracking and path fields are different from potential fields in that they do not nec-
following for marine vehicles [6]. Similar to this work is that essarily represent the gradient of a potential. Rather, the vector
of Skjetne et al. [7], which develops an output maneuvering field simply indicates a desired direction of travel. In [11], the
method composed of two tasks: forcing the output to converge concept of vector fields is used to direct autonomous aircraft in
to the desired path and then satisfying a desired speed assign- the stand-off tracking of a moving target. Precision tracking of
ment along the path. The method is demonstrated using a ma- a predefined path is not considered. Instead, the vector field is
rine vessel simulation. Reference [8] presents a path-following utilized to direct a team of two aircraft into an orbit around a
method for UAVs that provides a constant line of sight between moving target.
the UAV and an observation target. This paper considers the navigation of a fixed-wing MAV
The study presented in this paper builds on the concept of path with the assumption that altitude and airspeed are held con-
following through the construction of vector fields surrounding stant (or nearly so) by the control of the longitudinal dynamics.
the path to be followed. The vectors of the fields provide course
The following is a simple model of the navigational dynamics
commands to guide the MAV toward the desired path. As with
that will be used to study the path following behavior of the pro-
other path-following methods, the objective is not to track a
posed approach:
moving point, but to get onto the path while flying at a pre-
scribed airspeed. A unique contribution of this paper is the uti- (1)
lization of course measurements in the path-following control
(2)
which, in combination with the vector field strategy, guarantees
that tracking errors asymptotically approach zero even in the where represent the and components of the wind
presence of constant wind disturbances. Implementation of the velocity. Heading will be controlled by the vector-field
approach is feasible on small MAVs and experimental results path-following approaches presented in this paper. An alter-
validate the potential value of the approach for MAVs flying in native representation of these equations can be developed by
windy conditions.
noting the relationship between groundspeed , airspeed
, and wind speed as depicted in Fig. 2 as
II. PROBLEM DESCRIPTION
(3)
The objective of this paper is to develop a method for accurate
path following for MAVs in the presence of wind. The method (4)
calculates a vector field around the path to be tracked. The vec- Drawing on (3) and (4) and the definition of course shown
tors in the field are directed toward the path to be followed and in Fig. 2, (1) and (2) can be expressed as
represent the desired direction of flight. The vectors in the field
serve as course commands to the MAV. The method is currently (5)
applicable to paths composed of straight lines and arcs. This re- (6)
striction is insignificant for most practical applications. Fig. 1
shows examples of vector fields for linear and circular paths. The key distinction is that the equations of motion are expressed
The notion of vector fields is similar to that of potential fields, in terms of groundspeed and course and are independent of the
which have been widely used as a tool for path planning in the wind velocity. We will show that, by using ground-referenced
robotics community (see, e.g., [9]). It has also been suggested measurements (i.e., course and groundspeed instead of heading
in [10] that potential fields can be used in UAV navigation for and airspeed) in conjunction with the vector field approach to
obstacle and collision-avoidance applications. The method of control the path of the vehicle, wind-disturbance rejection is
[10] provides a way for groups of UAVs to use the gradient improved significantly, which is vitally important for small low-
NELSON et al.: VECTOR FIELD PATH FOLLOWING FOR MINIATURE AIR VEHICLES 521

Fig. 3. Vector field for straight-line path following. Far away from the waypoint
path, the vector field is directed with an angle  from the perpendicular to the
path.

Fig. 2. Relationship between the airspeed V , the windspeed W , and ground

speed V , as well as the relationship between heading and course .

speed MAVs. We will assume that the MAV is equipped with

an autopilot that implements a course-hold loop and that the
resulting dynamics are represented by the first-order system

(7)

where is the commanded course and is a known positive

constant that characterizes the speed of response of course-hold
autopilot loop.
In the development and analysis of the vector field approach
that follows, two primitive path types are considered: straight
lines and circular orbits. Circular arcs are treated similarly to
complete orbits. The approach is easily extended to paths com-
posed of multiple segments of arcs, orbits, and straight lines.
Fig. 4. Vector fields for various values of k . Large values of k yield abrupt
transitions from  to zero, while small values of k give smooth transitions.
III. TECHNICAL APPROACH

A. Straight Path Following

If is restricted to be in the range , then
Consider the straight-line path shown in Fig. 3. To follow clearly
this path, a desired-course vector field is constructed. Let be
the lateral distance of the MAV from the path and let be the
difference between the direction of the path and the course of
the MAV. Our objective is to construct the vector field so that,
when is large, the MAV is directed to approach the path with
course angle and that, as approaches zero, the course for all values of . Therefore, since is an odd func-
also approaches zero. Toward that end, define the desired course tion and is odd over , we can use the
of the MAV as Lyapunov function to argue that, if
, then asymptotically. Evaluating the Lie derivative
(8) of under the assumption that gives

where is a positive constant that influences the rate of the

transition from to zero. Fig. 4 shows how the choice of
affects the rate of transition. Large values of result in short,
abrupt transitions, while small values of cause long, smooth
transitions in the desired course. which is less than zero for .
522 IEEE TRANSACTIONS ON ROBOTICS, VOL. 23, NO. 3, JUNE 2007

In this paper, we use a sliding mode approach to render the

set

positively invariant and to ensure that the system trajectory

reaches in finite time. Let and differentiate to
obtain

(9)

Let and take the derivative to obtain

Fig. 5. State trajectories onto the sliding surface  (y ) for various initial
conditions.

shape of the trajectories onto the sliding surface. Large values

If we choose the control signal as of drive to zero quickly. The parameter is the width of
the transition region around the sliding surface that is used to
(10) reduce chatter in the control.
Theorem III.1: The system of equations given by (6) and (9),
where is given by (12) and is given by (8) is globally
where
exponentially stable if and
if
if (13)
if

and , then where

(11)

from which we conclude that in finite time [12].

It is well known that the sign function leads to chattering in and is arbitrary.
the control [12]. To mitigate the adverse effects of chattering, The stability condition (13) makes sense physically. If is
the control signal (10) is replaced with large, than the turning radius of the MAV increases, making it
more difficult to remain within the boundary layer. Similarly,
small or small imply slower convergence toward the path,
as indicated by Figs. 3 and 4. Small makes the feedback control
(12) high gain which enhances tracking. It is interesting to note that
(13) is independent of . However, from (12), it is clear that, to
avoid unrealistic control effort, is required to be proportional
where
to . Therefore, (13) is easier to satisfy if is large, implying
if that the on-board autopilot can quickly track course changes.
otherwise Proof: If , then is equivalent to (10), which
results in (11), implying that the set
and defines the width of the boundary region around the
sliding surface.
The shape of the sliding surface and system trajectory onto
the sliding surface are influenced by the control parameters is positively invariant and that converges to in finite time.
, and . Fig. 5 shows state trajectories onto the sliding It remains to show that, inside , the system trajectories con-
surface for various initial conditions generated from (6) verge to the origin . Toward that end, define the
–(8), and (12). The parameter controls the slope of the sliding Lypunov function candidate
surface near the origin, while scales the magnitude of
for large values of . The gain parameter controls the
NELSON et al.: VECTOR FIELD PATH FOLLOWING FOR MINIATURE AIR VEHICLES 523

where is a positive scale factor used to weight the terms so for , then for . Note
that they are similar in magnitude. Differentiating, we obtain that when

Inside the boundary region, we have

Therefore

if

Also note that, when , we have

Noting that

which implies that if

we get which is equivalent to

Therefore, if

Let

then which implies that

and note that for small values of and

for large . Consider the function Therefore, for , we get
if
if

where is arbitrary. Our objective is to find that ensures that (14)

.
To do this, note that both and are symmetric functions which is negative definite if
in . Therefore, without loss of generality, we will restrict our
attention to . To show that , we will use
the fact that if two functions and satisfy and
524 IEEE TRANSACTIONS ON ROBOTICS, VOL. 23, NO. 3, JUNE 2007

where is a constant that specifies the rate of transition

from to . This expression for is valid for all
values of .
For orbit following, it is convenient to change the navigational
dynamics to polar coordinates in terms of and , where the
center of the orbit is the origin. From Fig. 6,
and , where is the center of the orbit.
Taking the derivative and substituting into (5) and (6) gives

(16)
(17)

where the and are the ground track speed and relative
Fig. 6. Vector field geometry for orbit tracking. When the radial distance to the
MAV is much greater than the orbit radius, the desired course is calculated so course, respectively. We will again assume that the course dy-
that the MAV is directed toward the orbit center. As the radial distance becomes namics are given by
smaller, the desired course becomes tangential to the orbit.
(18)

Exponential stability comes from the fact that and the right- Defining , then we can argue that asymp-
hand side of (14) are quadratic [12]. totically when by using the Lyapunov function
For , we get , whose Lie derivative is

is less than zero for since the argument of is in the

set for all , implying that asymptotically.
We will again use a sliding-mode approach to render the set

which is less than zero if positively invariant and to ensure that the system trajectory
reaches in finite time. As before, define and
differentiate to obtain

thus implying asymptotic stability for .

B. Orbit Following
The algorithm for circular orbits creates vector fields in a
manner similar to the straight-line algorithm. Consider the de- where
sired orbit path shown in Fig. 6. In this discussion, a counter-
clockwise orbit will be considered. The development for clock-
wise orbits is similar with the exception of several sign changes.
The desired orbit is assumed to have a known center and radius has been defined for brevity. Letting gives
. When the distance between the MAV and the center of the
orbit is large, it is desirable for the MAV to fly toward the orbit
center. If we define as the radial distance of the MAV from
the center of the orbit, then, when is significantly larger than If we choose the control signal as
, the desired course is

(19)
where is defined as the angular position of the MAV with
respect to the orbit center, as shown in Fig. 6. where , then
When , the desired course is . Therefore,
let the desired course be given by

(15) from which we conclude that in finite time [12].

NELSON et al.: VECTOR FIELD PATH FOLLOWING FOR MINIATURE AIR VEHICLES 525

To avoid chattering in the orbit case, we replace (19) with

(20)

where defines the width of the boundary region around

the sliding mode.
Theorem III.2: The system of equations given by (16) and
(18), where is given by (20) and is given by (15), is
globally exponentially stable if

(21)

where

and is arbitrary.
Proof: The proof follows a similar line of reasoning as the
proof of Theorem III.1. If , then the set
Fig. 7. Straight-line paths and circular orbits can be combined to produce a
is positively invariant and converges to in finite time. Inside variety of paths. The top figure shows the combination of two straight-line paths
the boundary region, we use the Lyapunov function candidate and three circular orbits that are arranged so that, if the MAV is on the path, the
path length is equal to the original waypoint path. The middle figure arranges
the orbits so that the desired path transitions over the waypoint. In the bottom
figure, a single orbit is used to transition between straight-line segments.

with to obtain
For , we obtain

where . Following steps that are similar to

the proof of Theorem III.1, we obtain which is if

(23)

The combination of (22) and (23) leads to (21).

Letting
C. Combining Straight Lines and Orbits
and Many paths planned for MAV flight can be approximated by
if combinations of straight-line segments and circular arcs [13].
if Fig. 7 shows how combined paths can be utilized with waypoint
planning to fly paths that preserve equal path lengths, fly directly
over the waypoints, or turn in order to minimize flight time.
where is arbitrary, we can show that if
There are also a number of other situations where a combination
. From this, we find that for
would be desirable. For example, following a perimeter with
irregular geometry could be done effectively by approximating
its geometry with a series of lines and arcs.
When combining straight and arc segments, an approach for
constructing the vector field must be developed. In order to
avoid the possibility of multiple sinks, dead zones, and singular-
which is negative definite if ities that are inherent in the combination of vector fields, only
the vector field for the current segment or arc to be followed is
calculated. For a multisegmented path, the vector field changes
(22)
each time the MAV reaches the end of a segment or arc. Once
526 IEEE TRANSACTIONS ON ROBOTICS, VOL. 23, NO. 3, JUNE 2007

TABLE I
CONTROL PARAMETER VALUES

User interaction with the MAV is enabled by the Virtual

Cockpit software that interfaces through a communication box
to the MAVs. Virtual Cockpit allows the user to control the
behavior of the MAV by changing control gains, specifying
Fig. 8. Kestrel autopilot system developed at BYU.
waypoints, or changing experimental objectives. Telemetry
from the MAV is displayed in the Virtual Cockpit in real time,
allowing the user to monitor the progress of the experiment.
To implement the straight-line and orbit tracking control
laws, several control parameters must be specified. Table I
specifies the parameters used for the experimental results
presented. With these values, the stability conditions specified
by (13) and (21) are satisfied provided that is selected to scale
the following and course errors so that they are of similar mag-
nitudes during path following. For example, in the straight-line
case, choosing so that

Fig. 9. Zagi foam flying wing airframe.

ensures that the stability condition is a meaningful indicator of

the MAV has reached the end of a segment or arc, the entire system behavior.
vector field changes to direct the MAV onto the next segment
or arc. No two fields are combined, thus eliminating any issues B. Experimental Results
related to the combining of fields. To demonstrate the path-following abilities enabled by the
The method for determining when to change the vector field vector field algorithm, MAVs were commanded to fly a variety
must be specified. There are a number of methods for doing this. of paths composed of straight lines, orbits, and combinations
One way is to detect when the MAV is within a predetermined of straight lines and circular arcs. Experiments were conducted
distance from the end of the segment or arc. This works well for on three separate days. The majority of the results (which are
transitioning out of a straight path segment. For transitioning out presented first) were collected on a relatively calm day with the
of an arc, monitoring the angular travel of the MAV has proven average wind speed measuring 0.9 m/s from 230 southwest.
to be successful. Using this approach, the MAV transitions to The commanded airspeed was 13 m/s, thus for these tests the
the next path segment when the angle through which the MAV average wind speed was 7% of the commanded airspeed. Results
has flown is equal to the included angle of the arc. were also gathered on two high-wind days: one with an average
wind speed of 9 m/s and another with an average wind speed of
IV. RESULTS AND DISCUSSION 6 m/s.
The position of the MAV was measured using the on-board
A. Hardware Testbed GPS unit. The bias error associated with the GPS measurement
BYU has developed a reliable and robust platform for testing is approximately 10 m and is virtually unchanging over the du-
unmanned air vehicles [14]–[16]. Figs. 8 and 9 show two key ration of the flight experiments. Random errors are in the range
elements of the testbed. Fig. 8 shows Brigham Young Univer- of 2–3 m. The path-following errors reported are the calculated
sity’s (BYU) Kestrel autopilot which is equipped with a Rabbit by computing the lateral distance between the desired path and
3400 29-MHz processor, rate gyros, accelerometers, and abso- the location of the MAV as measured by GPS.
lute and differential pressure sensors. The autopilot measures To illustrate orbit following with the vector field algorithm,
3.8 5.1 1.9 cm and weighs 17 grams. the MAV was commanded to fly a series of concentric orbits
Fig. 9 shows the airframe used for the flight tests reported with varying radii. The results are shown in Fig. 10. The mean
in this paper. The airframe is a 1.2-m-wingspan Zagi XS EPP lateral path error for the 150-m orbit was 0.58 m, while the stan-
foam flying wing, which was selected for its durability, ease dard deviation of the path error was 0.42 m. For the 100-m orbit,
of component installation, and flight characteristics. Embedded the mean path error was 0.65 m and the standard deviation of the
in the airframe are the Kestrel autopilot, batteries, a 1000-mW path error was 0.46 m. For the 70-m orbit, the mean and standard
900-MHz radio modem, a GPS receiver, a video transmitter, and deviation of the lateral error was 1.9 and 1.1 m, respectively. For
a small analog camera. the larger orbits, the mean errors are about half of the wingspan
NELSON et al.: VECTOR FIELD PATH FOLLOWING FOR MINIATURE AIR VEHICLES 527

Fig. 10. Telemetry plot for orbits with radii of 150, 100, and 70 m. Fig. 12. Telemetry plot for equal path-length following.

Fig. 11. Telemetry plot for straight-line following. Fig. 13. Combination of equal path length and corner cutting following.

of the MAV. The data indicate that tight tracking is more dif- the straight-line and smoothed path lengths. The mean path-fol-
ficult to achieve as the orbit radius decreases. This is expected lowing error and standard deviation were 3.6 and 5.1 m, respec-
since control adjustments for loss in altitude in turns must be tively. Although the transitions from the straight line to the orbit
mixed with those used to control the lateral error and since the portions show some lateral following errors, the length of the
states associated with a smaller orbit are further away from the path flown and the desired length are similar. The length of the
nominal wings-level trim condition. straight-line path shown in Fig. 12 was 2897 m. The actual dis-
Fig. 11 illustrates the ability of the MAV to follow straight tance flown was 27 m less than the desired distance, which is an
line segments with acute angles. Excluding the transient errors error of only 0.93%.
from the turns, the mean following error on the straight-line To further test the robustness and capabilities of the proposed
portions of the path was 0.74 m with a standard deviation of path-following algorithms, many other types of paths have been
0.66 m. flown. The path shown in Fig. 13 illustrates both obtuse and
A combination of the straight-line and arc-following methods acute angles and the decision of the path follower to cut the cor-
was also tested. The techniques described in Section III-C were ners of the obtuse angles and flare out and around on the acute
implemented, and the results are plotted in Fig. 12. The thicker angles. Considering the path-following error over the full path
line represents the desired path that was planned to equalize gave a mean error of 3.6 m and a standard deviation of 4.7 m.
528 IEEE TRANSACTIONS ON ROBOTICS, VOL. 23, NO. 3, JUNE 2007

Fig. 14. Urban terrain following using straight-line following. Fig. 16. Straight-line following in high-wind conditions. Wind speed was 46%
of commanded airspeed.

Straight-line path-following results gathered under high-wind

conditions are shown in Fig. 16. The wind was measured at 6 m/s
coming from the south. The commanded airspeed was 13 m/s.
As expected, the largest errors occur during the down-wind por-
tions of a turn. In these situations, the groundspeed of the MAV
is high, and the minimum turn radius, which is governed by
roll-angle limits, becomes larger. It can be seen that, on the
straight-line portions of the path, the following errors are small.
In this case, the mean following error was 0.5 m, while the stan-
dard deviation of the error was 1.2 m along the straight portions
of the path.

V. CONCLUSION
In this paper, a new method for MAV path following, based
on the concept of vector fields, has been introduced. Using
Lyapunov stability criteria, it has been shown that the vector
field approach provides asymptotic following for straight-line
Fig. 15. Orbit following in high-wind conditions. Wind speed was 50% of com- and circular paths in the presence of constant wind disturbances.
manded airspeed. The effectiveness of the vector field method has been demon-
strated experimentally using a fixed-wing MAV. For both
straight-line and circular paths, following errors averaged less
Fig. 14 shows a path representative of an urban scenario. Al- than one wingspan in steady straight-line or orbit paths and less
though these are actual flight results, the terrain is synthetic. The than three wingspans for paths involving frequent transitions
straight-line follower was used to follow this path. The mean between straight-line and arc segments.
lateral error over the full path was 3.4 m, while the standard de-
viation was 4.8 m. REFERENCES
To demonstrate the feasibility of the approach for high-wind
[1] S. Park, J. Deyst, and J. How, “A new nonlinear guidance logic for tra-
conditions, an orbit path and a straight-line path were flown on jectory tracking,” in Proc. AIAA Guidance, Navigation Control Conf.,
days where the average wind speeds were 9 and 6 m/s, respec- Aug. 2004, paper no. AIAA-2004-4900.
tively. A plot of the orbit results is shown in Fig. 15. For these [2] I. Kaminer, A. Pascoal, E. Hallberg, and C. Silvestre, “Trajectory
tracking for autonomous vehicles: An integrated approach to guidance
results, the high wind required an airspeed of 18 m/s to enable and control,” AIAA J. Guidance, Control Dynam., vol. 21, no. 1, pp.
the MAV to make steady progress upwind. The wind speed was 29–38, 1998.
approximately 50% of the commanded airspeed. Under these [3] P. Aguiar, D. Dačić, J. Hespanha, and P. Kokotivić, “Path-following
or reference-tracking? An answer relaxing the limits to performance,”
conditions, the mean following error was 0.5 m with a standard in Proc. 5th IFAC/EURON Symp. Intell. Autonomous Vehicles, Lisbon,
deviation of 4.2 m. Portugal, 2004.
NELSON et al.: VECTOR FIELD PATH FOLLOWING FOR MINIATURE AIR VEHICLES 529

[4] P. Aguiar, J. Hespanha, and P. Kokotović, “Path-following for nonmin- D. Blake Barber received the B.S. degree in
imum phase systems removes performance limitations,” IEEE Trans. mechanical engineering from Brigham Young
Automat. Control, vol. 50, no. 2, pp. 234–239, Mar. 2005. University (BYU), Provo, UT, in 2005, where he is
[5] J. Hauser and R. Hindman, “Maneuver regulation from trajectory currently working toward the M.S. degree.
tracking: Feedback linearizable systems,” in Proc. IFAC Symp. Non- He has been actively involved in researching and
linear Control Syst. Design, Tahoe City, CA, Jun. 1995, pp. 595–600. testing new flight control algorithms and sensor plat-
[6] P. Encarnaçã and A. Pascoal, “Combined trajectory tracking and path forms for miniature UAVs. His research includes the
following: An application to the coordinated control of marine craft,” in use of optic flow sensors and vision algorithms for
Proc. IEEE Conf. Decision Control, Orlando, FL, 2001, pp. 964–969. range measurement, autonomous landing, path fol-
[7] R. Skjetne, T. Fossen, and P. Kokotović, “Robust output maneuvering lowing, and target localization.
for a class of nonlinear systems,” Automatica, vol. 40, pp. 373–383,
2004.
[8] R. Rysdyk, “UAV path following for constant line-of-sight,” in Proc.
2nd AIAA Unmanned Unlimited Conf., 2003, paper no. AIAA-2003-
6626. Timothy W. McLain (S’91–M’93–SM’03) received
[9] O. Khatib, “Real-time obstacle avoidance for manipulators and mo- B.S. and M.S. degrees from Brigham Young Univer-
bile robots,” in Proc. IEEE Int. Conf. Robot. Autom., 1985, vol. 2, pp. sity (BYU), Provo, UT, in 1986, and 1987, respec-
500–505. tively, and the Ph.D. degree from Stanford University,
[10] K. Sigurd and J. P. How, “UAV trajectory design using total field col- Stanford, CA, in 1995, all in mechanical engineering.
lision avoidance,” in Proc. AIAA Guidance, Navigation Control Conf., He is currently an Associate Professor with the
Aug. 2003, paper no. AIAA-2003-538. Mechanical Engineering Department, BYU. He
[11] E. Frew and D. Lawrence, “Cooperative stand-off tracking of moving has been actively involved in the control of air and
targets by a team of autonomous aircraft,” in Proc. AIAA Guidance, underwater vehicles and robotic systems for the past
Navigation Control Conf., Aug. 2005, paper no. AIAA-2005-6363. 18 years. During the summers of 1999 and 2000, he
[12] H. K. Khalil, Nonlinear Systems, 3rd ed. Upper Saddle River, NJ: was a Visiting Scientist with the Air Force Research
Prentice-Hall, 2002. Laboratory. Since that time, he has pursued research involving the modeling
[13] E. Anderson, R. Beard, and T. McLain, “Real-time dynamic trajec- and control of miniature UAVs, real-time trajectory generation for UAVs, and
tory smoothing for unmanned air vehicles,” IEEE Trans. Control Syst. cooperative control of UAV teams.
Technol., vol. 13, no. 3, pp. 471–477, May 2005.
[14] The Kestrel Autopilot. Procerus Technologies. [Online]. Available:
http://procerusuav.com/
[15] R. Beard, D. Kingston, M. Quigley, D. Snyder, R. Christiansen, W. Randal W. Beard (S’91–M’92–SM’03) received the
Johnson, T. McLain, and M. Goodrich, “Autonomous vehicle tech- B.S. degree in electrical engineering from the Uni-
nologies for small fixed wing UAVs,” AIAA J. Aerosp., Comput., Inf., versity of Utah, Salt Lake City, in 1991, and the M.S.
Commun., vol. 2, no. 1, pp. 92–108, Jan. 2005. degree in electrical engineering, the M.S. degree in
[16] R. W. Beard, D. Lee, S. Thakoor, and S. Zornetzer, “A new approach mathematics, and the Ph.D. degree in electrical engi-
to observation of descent and landing of future Mars mission using neering from Rensselaer Polytechnic Institute, Troy,
bioin-spired technology innovations,” AIAA J. Aerosp., Comput., Inf., NY, in 1993, 1994, and 1995, respectively.
Commun., vol. 2, no. 1, pp. 65–91, Jan. 2005. Since 1996, he has been with the Electrical and
Computer Engineering Department, Brigham Young
Derek R. Nelson received the B.S. and M.S. degrees University, Provo, UT, where he is currently an Asso-
from Brigham Young University, Provo, UT, in ciate Professor. In 1997 and 1998, he was a Summer
2004 and 2005, respectively, both in mechanical Faculty Fellow at the Jet Propulsion Laboratory, California Institute of Tech-
engineering. nology, Pasadena. In 2006 and 2007, he was a Visiting Research Fellow with
He is currently a Guidance and Controls Engineer the Air Force Research Laboratory, Munitions Directorate, Eglin AFB, FL. His
working on the Global Hawk Program at Northrop primary research focus is autonomous control of miniature air vehicles and mul-
Grumman Corporation, San Diego, CA. At BYU, he tivehicle coordination and control. He is currently an Associate Editor for the
was involved with UAV research, focusing on path IEEE Control Systems Magazine and the Journal of Intelligent and Robotic
following and cooperative control algorithms. Systems.