I.

INTRODUCTION

UAV Path Planning with Tangent-plus-Lyapunov Vector Field Guidance and Obstacle Avoidance
HONGDA CHEN KUOCHU CHANG George Mason University CRAIG S. AGATE Toyon Research Corporation

A dynamic path-planning algorithm is proposed for routing unmanned air vehicles (UAVs) in order to track ground targets under path constraints, wind effects, and obstacle avoidance requirements. We first present the tangent vector field guidance (TVFG) and the Lyapunov vector field guidance (LVFG) algorithms. We demonstrate that the TVFG outperforms the LVFG as long as a tangent line is available between the UAVs turning circle and an objective circle, which is a desired orbit pattern over a target. Based on a hybrid version of the TVFG and LVFG, we then derive a theoretically shortest path algorithm with UAV operational constraints given a target position and the current UAV dynamic state. This algorithm has the efficiency of the TVFG when UAV is outside the standoff circle and the ability to follow the path via the LVFG when inside the standoff circle. In addition we adopt point-mass approximation of the target state probability density function (pdf) for target motion prediction by exploiting road network information and target dynamics as well as obstacle avoidance strategies. Overall, the proposed technical approach is practical and competitive, supported by solid theoretical analysis on several aspects of the algorithm performance. With extensive simulations we show that the tangent-plus-Lyapunov vector field guidance (T+LVFG) algorithm provides effective and robust tracking performance in various scenarios, including a target moving according to waypoints or a random kinematics model in an environment that may include obstacles and/or winds. Manuscript received March 18, 2011; revised October 7, 2011, and January 27 and March 29, 2012; released for publication June 14, 2012. IEEE Log No. T-AES/49/2/944504. Refereeing of this contribution was handled by W. Koch. Authors current addresses: H. Chen, Sigma Space Corporation, 4600 Forbes Blvd., Lanham, MD 20706; K. Chang, Department of Systems Engineering and Operations Research, George Mason University, MS 4A6, 4400 University Dr., Fairfax, VA 22030, E-mail: (kchang@gmu.edu); C. S. Agate, Algorithms R&D, Toyon Research Corporation, Goleta, CA 93117.

c 2013 IEEE 0018-9251/13/$26.00 840

Path planning for multiple unmanned air vehicles (UAVs) to cooperatively track ground targets is an important research topic with approaches varying from classical optimal trajectory planning to bio-inspired swarm behavior [14]. There are many complex factors involved in creating a realistic path, such as obstacles, threats, the presence of multiple UAVs, uncertainty in the environment, and multiple tasks per UAV. The path-planning problem for a UAV tracking a ground target was typically formulated as an optimal control problem [59], consisting of a system dynamic, a set of boundary conditions, control constraints, and an object function. Traditionally, the objective function can be expressed as the cost associated with the UAV state trajectory along the path from a UAVs initial state to a desired state. The objective function is to be minimized in order to produce an optimal control solution. However, the overall costs will be UAV dependent and could include factors such as power (battery, gas) consumption, communication, mission scheduling with multiple UAVs, and so on. Past approaches include nonlinear programming [1012], optimization and stochastic sampling [1314], search methods [1519], and evolution based approaches [2021]. Assuming that each UAV has the resources to complete its tracking assignment, the cost function could be dramatically simplified. In this paper the only physical limitations we consider are a UAVs maximum turn rate and a nominal cruising speed. Consequently, the simplified objective considered here is to minimize the UAVs flight path length, such that given a UAVs initial state and a desired final state, the objective is to find the shortest trajectory between them with possible constraints. The focus of the paper is to develop a dynamic path-planning algorithm for routing UAVs to track ground targets under path constraints, wind effects, and obstacle avoidance requirements. In the case of obstacle-free environments, the control solution of the objective generally is a set of UAV waypoints and speeds passed to a UAV autopilot system (AP) for low-level control [2223]. Past approaches for trajectory planning include the map-based approach [2425], target-based approach with possible coordination between multiple UAVs [2629], and coordinated planning under potential uncertain environments [3032]. Considering UAVs limitations of turning rate and speed constraints, an accurate UAV motion model is required to insure that its AP can reach the waypoints. The shortest path involves flying a constant turning rate path. If we consider varying the UAV speed, a shorter distance solution is available such that UAV initially slows to achieve maximum turning rate, then accelerates along

IEEE TRANSACTIONS ON AEROSPACE AND ELECTRONIC SYSTEMS VOL. 49, NO. 2 APRIL 2013

tangent line. Finally, UAV decelerates on final circle to achieve desired velocity. In reality often there are obstacles or constraints such as a no-flight zone to restrict UAV flight path [3337]. Vision-based navigation and guidance strategy for collision avoidance are traditionally used in the literature [4448]. In this paper we develop a novel UAV routing algorithm with an obstacle avoidance strategy. The algorithm combines the tangent vector field guidance (TVFG) [38] path-planning approach with the Lyapunov vector field guidance (LVFG) algorithm [3941] to obtain a theoretically shortest path with UAV operational constraints given a target position and the current UAV dynamic state. This hybrid algorithm, called tangent-plus-Lyapunov vector field guidance (T+LVFG), has the advantage of the TVFG when UAV is outside the standoff circle and presents the ability to follow the path via the LVFG when inside the standoff circle. Given the assumption that the UAV has been given the initial state (position/velocity) information about the target of interest and that the UAV could observe the latest target state information with its own sensor in a tracking stage, we first determine the approximate location of target at intercept time. Then, a desired UAV state (at the planning horizon time) is chosen based on three factors: 1) UAV position is located on the circle of a predefined radius about target; 2) UAV speed is selected based on the target speed (matched if possible); and 3) UAV heading is the same to the target. Usually, the estimated intercept location is less accurate when UAV is farther away. However, the replanning interval is normally shorter than the planning horizon so that accuracy of prediction improves as UAV approaches target. Moreover, considering target location uncertainty or probability of detecting target at a desired location, we propose to use a particle filter to predict target location probability density function (pdf). A point-mass approximation of the target state pdf is stochastically simulated using a ground vehicle motion model that exploits road network and other information. Once we generate a predicted target state pdf, we determine an optimal UAV location that maximizes the probability of detecting the target. When obstacles or constraints are present to restrict the UAV flight path, we model the obstacles as a circular region similar to the standoff circles. The obstacle avoidance strategy is then developed by considering the path with obstacles and determining the best path to go around them with the similar TVFG concept mentioned earlier. Overall, the main contribution of this research is the newly proposed path-planning method, which is able to provide a shortest path for a UAV tracking a target, to utilize particle filters for target motion predictions and to avoid identified obstacles in generating the optimal paths.

The rest of this paper is organized as follows. In Section II the model and solution methodology of UAV flight guidance used to create the path plans in the search of an uncertain, dynamic, and windy environment [42] is reviewed. T+LVFG is described in detail. Section III introduces the particle approach for target motion prediction to maximize target detection probability. Target-location particles and how they are utilized in the stochastic simulations are discussed. The calculation of UAV coverage probabilities is derived analytically. In Section IV obstacle avoidance strategies are presented and analyzed. Section V gives a set of demonstrations to validate the techniques in multiple different environments. Finally, Section VI concludes the paper. II. UAV FLIGHT GUIDANCE

The UAV autopilot system is developed to accomplish more complex tasks, operate in uncertain and dynamic environments, and eventually allow integration with other UAVs to accomplish coordinated missions. Its command signal is produced by the guidance system with respect to UAV state variables. The algorithms presented here assume that the UAV is equipped with a low-level control system so that vector field guidance indications can present a kinematic model to the guidance layer. In an obstacle-free environment the dynamic path-planning algorithm requires the following information about the system state: 1) the predicted target state, 2) the current UAV state, and 3) the desired UAV state. Specifically, a two-dimensional (2D) position and velocity target state is defined by _ t , yt , y _ t ]; the UAV inertial state is [xu , x _ u , yu , y _ u ]; the [xt , x _ r , yr , y _ r ]; relative state of the UAV to the target is [xr , x _ d , yd , y _ d ]. It is and the desired UAV state is [xd , x necessary to mention that all state vectors are time dependent, however, the time variable (t) is not presented for simplicity. In practice the UAV inertial state at any time step is known, and the target state is estimated from collected sensor data. The desired UAV state is given by the following conditions: 1) the UAV is positioned on a circle of fixed radius around the estimated target location; 2) the UAVs heading is equal to the targets heading; and 3) the UAVs speed is equal to the targets speed unless the target speed is less than the UAVs minimal speed; in which case, the UAVs speed is set to its minimum. The UAVs desired state is, therefore, dependent on the target state and can be determined through different approaches. For example, in the left diagram of Fig. 1, UAV starts at point A with initial velocity v1 = _ u (t1 )] and ends at desired endpoint G with _ u (t1 ), y [x _ d (t2 ), y _ d (t2 )]. It is obvious that the size velocity v2 = [x of a UAVs turning circle depends on its speed. The optimal path (minimal traveling distance) follows A !
841

CHEN, ET AL.: UAV PATH PLANNING WITH TANGENT-PLUS-LYAPUNOV VECTOR FIELD GUIDANCE

Fig. 1. UAV path planning (L) and selected path (R).

B ! C ! D ! E ! F ! G, in which the BC circles correspond to the minimal turning radius (slowest speed), the large dotted circle (with the tangent line CD) corresponds to maximal UAV cruising speed, the AB circle corresponds to the initial UAV speed, and the EF circle corresponds to the final desirable standoff circle. The optimal path from A to G can be illustrated as in the right side of Fig. 1, in which the UAV first slows down to follow a minimum turning circle, then it speeds up to a maximum speed along a tangent line, and finally it slows down again to reach the desired standoff circle. Therefore, the optimal control problem can be described as: given the UAV position and velocity at points A and G, find the points B/C/D/E/F such that the cost function (travel distance) is minimized. In the subsequent section we show that the shortest path is found using what we call tangent vector guidance. A. Lyapunov Vector Field Guidance Using an LVFG law [4041], the guidance of a UAV to an observation orbit around a target can be determined by building a vector field that has a stable limit cycle centered around the target position. The UAV is assumed to be able to move freely but only in the direction of its orientation. As defined in [39] the UAV dynamics can be modeled as _ u = V cos ; x _ u = V sin ; y _ =w u (1)

Fig. 2. Clockwise LVFG with Rt = 30 m.

be nonpositive by choosing a desired relative vehicle _ d, y _ d ]T according to the guidance vector velocity [x field, such that [41] 2 _d r Rt2 xr 2rRt x = (2) 2 2 _d 2rRt r Rt yr y in an anticlockwise direction, or 2 _d r Rt2 2rRt xr x = _d 2rRt r2 Rt2 yr y

(3)

in a clockwise direction, where is a nonnegative normalization factor. With (2) and (3) the velocity field vectors produce a nonpositive rate of change of (). If is larger than zero, () is zero and invariant only on the standoff circle (r = Rt ), which ensures that the vector field produces a globally attractive limit cycle [41]. For example, Fig. 2 illustrates the guidance vectors surrounding a stationary target in the clockwise orientation. It is clear that a path following the field vectors from any point will end up on a circle of radius equal to the specified standoff distance (shown in black). B. Tangent Vector Field Guidance In [38] we proposed a dynamic path-planning algorithm for a UAV that was tracking a ground target. A theoretically optimal path based on the TVFG was derived with UAV operational constraints given a target position and the current UAV kinematic state. The limitation of TVFG is that the UAV must be outside the standoff circle; otherwise, a tangent line does not exist, and the TVFG cannot provide proper guidance. In general, for any 2D point (xu , yu ) outside of an objective circle with radius Rc and center (xc , yc ), we can easily calculate two tangent points on the circle,

where the inputs V and w denote lateral and angular velocity, whose constraints are given by vmin V vmax and w . A UAV traveling at constant speed without slipping will experience a minimum turning radius so that the angular velocity is bounded by jwj V=R . Given the two-dimensional inertial position of the aircraft [xu , yu ]T and its relative position to the target p = [xr , yr ]T , a Lyapunov vector field law can be applied to determine the guidance of a UAV by calculating the desired velocity. Consider the Lyapunov function (p) = (r2 Rt2 )2 in which p 2 2 r = xr + yr is the radial distance of the UAV from the target and Rt is the radius of the standoff circle. The total time derivative of (p) can be specified to
842

IEEE TRANSACTIONS ON AEROSPACE AND ELECTRONIC SYSTEMS VOL. 49, NO. 2 APRIL 2013

_ t, y _ t ]T , the UAV model target [xt , yt ]T with velocity [x may be simply modified as _ _ r = V cos + W x x xt _ _ r = V sin + W y y yt (7) (8)

Fig. 3. Clockwise TVFG with Rt = 30 m.

such that q 2 Rc Rc 2 R2 + x ( x x ) ( y y ) ru u c u c c c 2 2 ru ru q 2 Rc Rc 2 R2 + y y1,2 = 2 (yu yc ) 2 (xu xc ) ru c c ru ru p where ru = (xu xc )2 + (yu yc )2 . In that case, assuming that [xo , yo ]T is the tangent point on the standoff circle, the UAV direction angle can be determined as y yo : r = tan1 u xu xo x1,2 =

(4)

(5)

The desired relative vehicle velocity according to the guidance vector field is _d cos r x = 2 (6) _d sin r y where 2 is a nonnegative normalization factor. For example, Fig. 3 visualizes the TVFG vectors surrounding a stationary target with clockwise orientation. It is clear that a path following the field vectors from any point will end up on a circle of radius equal to the specified standoff distance (shown in black). Moreover, it is easy to show the following result. PROPOSITION 1 As long as the tangent lines are available between a UAVs turning circle and an objective circle, TVFG provides more efficient (in the sense of shorter path) path planning than LVFG. The proof of Proposition 1 is provided in Appendix I. Essentially, this proposition states that TVFG performs strictly better than LVFG when UAV is outside the target standoff circle. In the presence of a wind field, when expressed relative to a moving

where W x and W y are the components of the background (horizontal and vertical) wind velocity. To compare the performance of the two guidance laws in a more rigorous manner, we introduce distance and time efficiency measures to evaluate the UAV path-planning efficiency. The distance between UAV and target is denoted as d. The relative UAV heading direction to the target is denoted , and is the relative direction of Wind. The distance and time efficiency are then defined as Dx =d and T x V=d , respectively. Here, Dx denotes the UAV total flight distance to the objective circle, and T x is the total time is the average used to travel to the objective circle. V speed of UAV. Notice that, in the case of a UAV with constant speed, the time efficiency and distance efficiency are equivalent. Figure 4 plots two example trajectories when = 0 and = , respectively. The UAV speed is constant at 14 m/s, the turning limit is =6, and the sampling period is 1 s. The radius of the objective circle is 30 m, the target is located at (0, 0), and the initial UAV is at (100 m, 100 m). Note that, in general, TVFG provides more efficient waypoints to reach the objective orbit. Additionally, we observe that the LVFG cannot precisely reach the desired standoff circle. The offsets depend on UAVs velocity and sampling interval. Figure 5 plots the distance/time efficiency as a function of relative angle between 0 to . Here, in order to have a fair comparison, we use the LVFGs orbit as the objective for evaluation. Time and distance efficiency are identical since we assume that UAVs speed is constant. Notice that the TVFG performs about 40% better than the LVFG on the efficiency measurements. C. Tangent-Plus-Lyapunov Vector Field Guidance (T+LVFG)

As shown in the previous section, TVFG is more efficient than LVFG when UAV is outside the standoff circle. However, when UAV is inside the standoff circle, TVFG is not applicable due to the lack of a tangent line. Because of the limitation we propose a hybrid strategy to take advantage of both TVFG and LVFG, such that whenever UAV is outside the standoff circle, we use TVFG, and whenever UAV is inside the standoff circle, we use LVFG. We refer to this hybrid strategy as the T+LVFG algorithm. As shown in the following proposition, T+LVFG performs better than each of the two stand-alone algorithms due to its hybrid nature.
843

CHEN, ET AL.: UAV PATH PLANNING WITH TANGENT-PLUS-LYAPUNOV VECTOR FIELD GUIDANCE

Fig. 5. Efficiency performances of LVFG and TVFG.

The proof of Proposition 2 is provided in Appendix II. Figure 6 shows the ASBs with different sampling intervals for a UAV that is tracking a stationary target. We observe that the bias T+LVFG is about an order of magnitude smaller than that of LVFG. III. TARGET MOTION ESTIMATION AND MAXIMIZING TARGET DETECTION As discussed previously we base the UAVs desired location on the estimated location of the target. Choosing a target location estimate, such as the expected location, may be straightforward when the predicted target location pdf is uni-modal (e.g., Gaussian). However, in many practical ground target tracking scenarios, the ground vehicles move through a road network, which, due to the presence of road intersections, often results in predicted target pdfs that are multi-modal. Moreover, we need a basis for choosing the desired UAV location. Thus, we use a point-mass approximation of the target state pdf and use stochastic simulation to propagate the pdf using a ground vehicle motion model that exploits road network information. Once we generate a predicted target state pdf, we determine an optimal UAV location that maximizes the probability of detecting the target. With this desired UAV position, we can apply the algorithm discussed previously to generate the optimal UAV trajectory. A. Coverage in 2D Space We first discuss the coverage probabilities of a UAV and a target in a 2D space. This simplified case assumes that the UAV and target can move along any direction without physical limitations. Specifically, u, y u ) and that the assume that the UAV position is (x t, y t ). The heading directions are target position is (x

Fig. 4. (a) Example 1relative angle of UAV heading to target is 0 deg. (b) Example 2relative angle of UAV heading to target is 180 deg.

As a UAV tracks a target along a standoff circle, the average distance between the UAV trajectory and the standoff circle in steady-state is defined as the average steady-state bias (ASB). Due to the finite sampling rate, the UAV cannot fly along the circle exactly but reaches a steady state with an error between the objective circle and the UAVs actual flight path. In general we have the following result. and a PROPOSITION 2 With a nominal speed V sampling time (planning cycle) , when the UAV is tracking a stationary target along a standoff circle Rc , the ASB of T+LVFG and LVFG are given by: q 2 2 1) In T+LVFG ASB is bounded by Rt2 + 1 4V Rt ; q p 3 2) In LVFG ASB is given by q + q2 + p3 + q p 3 q q2 + p3 + 1 6 V Rt , 1 3 3 2 2 2 and p = 1 V where q = V 1 VR
2 1 3 Rt . 216 3 t 36

844

IEEE TRANSACTIONS ON AEROSPACE AND ELECTRONIC SYSTEMS VOL. 49, NO. 2 APRIL 2013

Fig. 7. Geometry illustration for two angle calculations.

depends on the distance between UAV and the target, namely q (12) d0 = (xt xu )2 + (yt yu )2 : If d0 Rc + Rt the target is out of the field of view (FOV) of the UAV so that the coverage probability is zero. If d0 Rc Rt the target is completely in the FOV of the UAV so that the coverage probability is one. In Fig. 7 angles 1 and 2 can be calculated by the law of cosines, such that 2 2 Rt2 Rc + d0 and 1 = 2 cos1 2Rc d0 2 2 2 1 Rt + d0 Rc 2 = 2 cos : 2Rt d0 Therefore, based on the determined angles, one can readily show the following result. The proof is omitted. PROPOSITION 3 Suppose that the UAVs coverage radius is Rc and that the target uncertainty radius is Rt . The coverage probabilities (Pr) can be expressed as
Fig. 6. (a) Objective bias using T+LVFG with UAV speed 14 m/s. (b) Objective bias using LVFG with UAV speed 14 m/s.

assumed to be and , respectively. Then, at the next time step, the UAV and target positions can be expressed as u + V xu = x u cos u + V yu = y u sin t + V xt = x t cos : t + V yt = y t sin

1) In case of Rc Rt we have 8 s1 p 2 R2 if d0 Rc > t < R 2 , t Pr = : > : 1 s2 , otherwise Rt2 2) In case of Rc < Rt we have 8 s p 1 2 > , if d0 Rt2 Rc > < Rt2 Pr = > R2 s > : c + 2 , otherwise Rt2 Rt2

(13)

(9)

(14)

(10)

The radius of the UAVs coverage is denoted to be Rc , and target uncertainty radius (say, 3-sigma boundary) is Rt . The intersection points of the two circles may be found from the solutions of the following equations,
2 (x xu ) + (y yu ) = Rc (x xt )2 + (y yt )2 = Rt2 : 2 2

1 2 2 where s1 1 2 [2 sin(2 )]Rt + 2 [1 sin(1 )]Rc and 1 1 2 2 s2 2 [2 sin(2 )]Rt 2 [1 sin(1 )]Rc .

B. General Coverage Probabilities In practice, due to the road and/or turning angle limitations, a UAV or a target cannot move freely to anywhere in their reachable circles and, therefore, the circular representation of the targets location uncertainty is insufficient. Our objective for choosing a desired UAV location is to maximize the probability
845

(11)

Obviously, in the case of no real solution from (11), the coverage probability is either one or zero, which

CHEN, ET AL.: UAV PATH PLANNING WITH TANGENT-PLUS-LYAPUNOV VECTOR FIELD GUIDANCE

of detection. Since the targets state is not known with certainty, we can only calculate the expected probability of detection Z D (x)p(x)dx (15) EfPd g =
s

Based on the UAVs FOV radius, the coverage probabilities can then be estimated as a ratio of the number of particles inside the UAVs FOV over the total number of particles Pr(j ) = 1X [dP !C (i, j ) RC ]: N
i

(17)

where s presents the sensor FOV, p(x) is the pdf of the target state, and D (x) represents a function that maps the target state into a probability of detection. Note that p(x) is obtained by extrapolating the targets kinematic state based on road network information. The idea is to use a digital road map to constrain the possible target locations. Monte Carlo samples (particles) can be generated to describe possible target states, and the integration part of (15) can be approximated by counting the number of particles in the respective areas. C. Coverage Rate with Particles As mentioned in the previous section, we apply particles to simulate possible target locations at the next time step in order to obtain a UAVs desired location. As described in the Introduction, it is assumed that the UAV has been given the initial state (position/velocity) information about the target of interest; and in a tracking stage, the UAV could observe the latest target state information with its own sensor. We further assume that digital road map information is available and that initial target information is available from at least one of the resources (ground moving target indicator (GMTI) or otherwise). After initialization the UAV must track the target using only sensor data collected by the UAVs onboard sensor. If the UAV cannot acquire the target on its own after the initial assignment, then it is highly likely that it will never track the target. In that case a separate initialization/reassignment process is needed. The initial pdf of a target state is generally given based on the assumption that the target is following the roads in a given map, and its velocities are uniformly distributed between maximal and minimal speed limits. Thus, particles will be randomly drawn from this pdf. These particles are used to predict the target state forward in time based on the road network. The desired UAV destination objective is a reachable position for the UAV within a given planning cycle that maximizes the probability of detecting the target, which amounts to the UAV location such that the maximum number of particles is within the UAVs FOV. Euclidean distances can be used to compare the destination objectives. Specifically, by assuming that there are N particles fPi (x, y )g and M candidate objectives fCj (x, y )g, we have the distance between particle i and candidate objective j , namely dP !C (i, j ) = jPi (x, y ) Cj (x, y )j:
846

To determine the most desirable UAV destination objective based on (16) and (17), we must sample the set of reachable UAV states in order to render the number of potential locations to consider into a finite number. To select the destination candidates, one simple idea is to define them uniformly in the UAV reachable region based on maximum and minimum UAV speeds. IV. OBSTACLE AVOIDANCE In an operational environment often there are obstacles or constraints, such as no-flight zone or tall buildings, to restrict the UAV flight path. The information about the existence of the obstacles is either known in advance (e.g., buildings) or acquired during flight (e.g., collision avoidance with other aircraft). In most real scenarios more than one obstacle might appear, so multiple obstacles are considered. We assume that the UAV is capable of receiving obstacle information at any path-planning cycle by either its own sensors or road map indications. Without loss of generality obstacles are modeled as circles determined by their center position and radius. Other shapes may be treated in a conservative manner by inscribing them in a larger circle, which thereby permits the obstacle avoidance problem to be formulated in terms of a simplified model of turning circles similar to the shortest path algorithm described earlier. The obstacle avoidance problem is, then, interpreted as the shortest path selection among all possible paths. Specifically, we first determine the closest obstacle for the UAV to move around by using a straight line to link the UAV and the destination circle about the target. Then, we start from this obstacle to locate the next possible obstacle, if any, between the target and current obstacle circle. By this recursive process all the possible paths and corresponding distances can be calculated to obtain a practical solution with the shortest distance. More specifically, during a path-planning cycle, assume L obstacle circles (no flight zones) appear unexpectedly. These obstacle circle parameters are assumed to be available, i.e., center coordinates [i ] [i ] [i ] , yo ) and radius Ro . For simplicity the obstacle (xo circles are assumed to be nonoverlapping in this paper. The intersection points of the obstacle circles and the path-line between the UAV and the target may be found from the solutions of the following

(16)

IEEE TRANSACTIONS ON AEROSPACE AND ELECTRONIC SYSTEMS VOL. 49, NO. 2 APRIL 2013

equations
[i ] 2 [i ] 2 [i]2 ) + (y yo ) = Ro , (x xo

i = 1, 2, : : : , L (18)

x xt y yt = : yu yt xu xt

If there is no real solution in (18) for an obstacle-i, it means that this obstacle will not affect the current path planning, and we may ignore it at this moment. Otherwise, based on the distance measurements between UAV and the intersection points, we can readily locate the closest obstacle, which is blocking the path (marked as the most effective obstacle or MEO), and initiate the obstacle avoidance algorithm to plan the UAV path around it. In general the UAV will take the shortest distance by following the line path connecting the start point and the target point. In the case that an MEO is located, the path-planning algorithm will follow a tangent line to this obstacle. There are three basic situations to be considered: 1) the MEO is independent, 2) the MEO is dependent, and 3) without effective obstacle. In the situation of without effective obstacles, path planning operates in a normal manner, and the obstacle avoidance algorithm is not needed. For independent MEO, as illustrated in Fig. 8(a), the path-line of x1 ! y1 crosses obstacle-A and obstacle-B; the first MEO can be identified as obstacle-A. The obstacle avoidance algorithm will select a path around obstacle-A first, such that x1 ! x2. Similarly, by the straight-line x2 ! y2, the subsequent MEO is identified as obstacle-B, and the obstacle avoidance algorithm will find a path to go around it. Therefore, the overall path could be obtained as x1 ! x2 ! x3 ! y3. The case of dependent MEO is illustrated in Fig. 8(b). As the path-line of x1 ! y1 crosses only obstacle-C, an MEO is identified as obstacle-C. The obstacle avoidance algorithm will select a path around obstacle-C, e.g., x1 ! x2. However, this new path will cross another obstacle-A, which will affect the current path plan. Therefore, adjustments of the path plan need to be made first to avoid obstacle-A. In reality we simply treat obstacle-A as a pseudo objective circle, and then the TVFG can be applied to guide the UAV to go around it. V. NUMERICAL RESULTS Multiple simulation scenarios are provided in this section to illustrate the performance of the newly proposed dynamic UAV path-planning algorithm for ground target tracking. First, we consider three examples of different target-moving models [8, 12]: 1) a target is moving with a kinematic model (nearly constant velocity) in 2D free space, 2) a target is moving within time-based waypoints, 3) a target is

Fig. 8. (a) MEO independent. (b) MEO dependent.

moving within speed-based waypoints. Then, we consider different wind environments to present its robustness for the UAV to track a stationary and/or moving target. After that we present detection probabilities using an example. Moreover, the particle simulation approach is employed in two different scenarios. Finally, the obstacle avoidance algorithm is validated as multiple obstacles appear in dynamic path generations. In the first scenario a UAVs initial position is assumed to be at [x0 = 300, y0 = 200] m and its height 100 m. It has a nominal speed of 14 m/s with a heading angle [54 deg (East is 0 deg)]. The UAV maximal turning rate is 30 deg/s. Both the sampling rate and planning cycle are in 1 s. Using T+LVFG Fig. 9 plots four cases as a target follows the kinematic model (Fig. 9(a) and (b)), speed-based waypoint model (Fig. 9(c)), and time-based waypoint model (Fig. 9(d)). In the kinematic model the t = initial target position is assumed to be at [x0
847

CHEN, ET AL.: UAV PATH PLANNING WITH TANGENT-PLUS-LYAPUNOV VECTOR FIELD GUIDANCE

Fig. 9. T+LVFG tracking performance in four different scenarios. (a) T+LVFG tracking: Target with random-moving model. (b) T+LVFG tracking with winds: Target with random-moving model. (c) T+LVFG tracking: Target with speed-waypoint model. (d) T+LVFG tracking with winds: Target with time-waypoint model.
t 200, y0 = 100] m with an initial speed of 5 m/s and a heading angle of 45 deg. The target maximal speed is 12 m/s, and the state process noise standard deviation is 0.1 m/s. In addition a constant wind of strength 5.0 m/s and an angle of 45 deg is present in scenarios (b) and (d). In the speed-based model the target has a constant speed in each line space (road section). Specifically, the moving segments are defined as pt (t0 ) ) pt (t1 ) ) pt (t2 ) ) pt (t3 ) ) pt (t4 ) ) pt (t5 ), where the target positions are located at [100, 120] m, [50, 20] m, [10, 0] m, [50, 102] m, [110, 150] m, and [300, 200] m, and the target speeds are 10 m/s, 8 m/s, 7 m/s, 2 m/s, 15 m/s, 12 m/s, and 14 m/s, respectively. In the time-based model the target arrives to these denoted points at times of 0 s, 8 s, 18 s, 37 s, 52 s, and 68 s then stops at the final position. Results show that the

T+LVFG algorithm provides an efficient and effective solution in each of the scenarios. To demonstrate the robustness of the T+LVFG algorithm under wind conditions, we now consider a case in which the UAV tracks a stationary target. Suppose that a UAVs relative-angle to the target is , turn-limit is 30 deg/s with a speed of 14 m/s. Figure 10 plots the UAV and target trajectories in different wind environments. In Fig. 10(a) the trajectories using LVFG and T+LVFG under a wind condition of w = 1:0 m/s and w = =4 are given. While in this case the T+LVFG yields similar paths with or without the presence of wind, the LVFG alone does not perform very well. In Fig. 10(b) the wind strength is increased to w = 2:0 m/s and the heading angle is =4. Notice that, while the T+LVFG algorithm has a slightly distorted path along

848

IEEE TRANSACTIONS ON AEROSPACE AND ELECTRONIC SYSTEMS VOL. 49, NO. 2 APRIL 2013

Fig. 10. Performance comparison with different wind scenarios. The relative-angle of UAV and target is , UAV with v = 14 m/s and turn-limit W = =6. (a) Constant winds: strength = 1:0 m/s and angle w = =4. (b) Constant winds: strength = 2:0 m/s and angle w = =4. (c) Constant winds: strength = 13:0 m/s and angle w = =4. (d) Constant winds: strength = 14:0 m/s and angle w = =4.

the objective circle, the LVFG could not keep up with the target in these mild wind cases. As shown in Fig. 10(c), at a high wind speed of 13.0 m/s, the T+LVFG algorithm struggles but manages to eventually locate the target within the FOV, though the trajectory fails to accurately follow the objective circle. Unfortunately, when the wind speed is the same or higher than the UAVs speed, the UAV cannot compensate for the wind and, therefore, cannot reach the objective circle as expected. A simulation example with the wind speed equal to the UAV speed limit of 14 m/s is shown in Fig. 10(d) in which both algorithms diverge. Next, we apply the newly proposed T+LVFG algorithm for a UAV that tracks a moving target under windy conditions. The UAV has the same parameters

as in the case of tracking a stationary target. Figure 11 shows the tracking results for a moving target with and without winds. In Fig. 11(a) the moving target has a constant velocity of 5 m/s, starts from (0, 0), and moves along a direction angle of t = =10. Since the UAV is much faster than the target, both algorithms demonstrate that the UAVs are circling along the target trajectory. However, the T+LVFG algorithm provides closer and more efficient tracks than the LVFG. In Fig. 11(b) we consider a faster target with a constant velocity of 12 m/s. The target starts from (0, 0) and heads east. Again, the path corresponding to the T+LVFG algorithm is generally closer to the target. In Figs. 11(c) and (d), winds are added with a speed of 2 m/s and 13 m/s, respectively. The
849

CHEN, ET AL.: UAV PATH PLANNING WITH TANGENT-PLUS-LYAPUNOV VECTOR FIELD GUIDANCE

Fig. 11. Performance comparison for a moving target in different wind conditions. (a) Without winds, Target speed is 5 m/s, and UAV speed is 14 m/s. (b) Without winds, Target speed is 12 m/s, and UAV speed is 14 m/s. (c) Constant winds ( = 2 m/s, w = =4), Target speed is 5 m/s, and UAV speed is 14 m/s. (d) Constant winds ( = 13 m/s, w = =4), Target speed is 5 m/s, and the UAV speed is 14 m/s.

T+LVFG algorithm works well, but the LVFG alone fails in both cases. Figure 12 illustrates simulation results using a point-mass approximation of the target predicted pdf and choosing a desired UAV location based on maximizing the probability of detecting the target. The two plots show the corresponding tracking results via the particle simulation strategy. In each planning step we generate 100 particles for the predicted target location and select 50 destination candidates uniformly in the UAVs reachable region. At each planning cycle we select the UAV destination based on maximizing the probability of detecting the target. Finally, we evaluate the obstacle-avoidance strategies in the dynamic path-planning algorithm. Suppose that there are three obstacles in a path-planning environment. The simulated obstacles
850

are defined as Obstacle-1 centered at [10, 40] m with a radius of 45 m, Obstacle-2 centered at [110, 30] m and radius 50 m, and Obstacle-3 centered at [68, 140] m and radius 50 m. On the left of Fig. 13, the initial position of UAV is fixed at [160, 140] m, and a heading of 50 deg. A target is following a road map. The optimal paths are plotted for different hypothetical target positions, where obstacles are successfully avoided in each planned path. On the right of Fig. 13, the same obstacles are presented with a specific target trajectory. A target is moving with a kinematic model in 2D free space [8]. The initial t t target position is assumed to be [x0 = 200, y0 = 100] m with an initial speed of 5 m/s and a heading of 45 deg. The targets maximal speed is 12 m/s, and the state process noise standard deviation is 0.1 m/s. The UAV is initially located at [500, 200] m

IEEE TRANSACTIONS ON AEROSPACE AND ELECTRONIC SYSTEMS VOL. 49, NO. 2 APRIL 2013

with velocity [14, 0] m/s and heading of 90 deg. The results show that the UAV is able to track the target successfully while avoiding the obstacles. VI. CONCLUSIONS In this paper we have proposed a dynamic UAV path-planning algorithm for tracking a ground target. The algorithm is characterized by a combination of TVFG and LVFG, a point-mass approximation of the target state pdf, and obstacle avoidance strategies. The key advantages of the proposed method are the ability to find the shortest path for a UAV tracking target, to utilize particle filters for target motion predictions for the case in which road network information is available, and to avoid identified obstacles in the path generations. Theoretical analysis shows that the T+LVFG algorithm outperforms the LVFG alone whenever the tangent lines are available between the UAVs turning limit circle and an objective circle. The steady-state objective biases of both the T+LVFG and LVFG algorithms are formally derived, as well as the coverage probability calculations. The fact that tangent lines and particle filters are combined to generate path candidates for UAV tracking makes this approach very effective. The proposed solution methodology can be easily embedded in a mission-planning strategy. This dynamic path planning uses fixed-increment time advance, which fits well with any heuristic solution algorithms. Simulation results demonstrate the efficiency and robustness of the new algorithm as a UAV tracks a target in a variety of different experimental scenarios, including different target moving models, wind environments, preknown road maps, and multiple obstacles along the way. Future work will focus on a collaboration of multiple UAVs in a multiple target environment.

Fig. 12. (a) Map-1: Tracking results with road constraints. (b) Map-2: Tracking results with road constraints.

Fig. 13. Path planning and tracking results with the obstacle avoidance strategy. CHEN, ET AL.: UAV PATH PLANNING WITH TANGENT-PLUS-LYAPUNOV VECTOR FIELD GUIDANCE 851

APPENDIX I PROOF
OF

PROPOSITION 1

Without loss of generality, we assume that a target is located at [0, 0], the radius of an objective circle is Rt , and a UAV position is given by [xr , yr ]. As long as the tangent lines are available between the UAVs turning limit circle and the objective circle, TVFG follows tangent lines to the standoff circle. Thus, the minimal distance between the target and the TVFG guidance line is always a constant of Rt . Next we show that the minimal distance between the target and the LVFG guidance line is larger than Rt and that it depends on the location of the UAV. Based on pthe LVFG (2, 3), at the UAV position 2 + y 2 , two possible directional lines are [xr , yr ], r = xr r either clockwise as y yr = 2rRt xr + (r2 Rt2 )yr (x xr ) (r2 Rt2 )xr + 2rRt yr (19)

Fig. 14. Illustration diagram for TVFG objective bias.

Figure 14 illustrates a UAV that passes through an objective circle via TVFG. A target is located at O, and the radius of the objective circle is R = Rt . Assume that the UAVs current position is A and that it will be moving along the tangent line A ! B with the tangent point C. If a planning cycle time is seconds and if the points A, B are associated with two consequent planning cycles, then we have jAB j = jAC j + jCB j = a1 + a2

or counterclockwise as 2rR x + (r2 Rt2 )yr y yr = 2 t r 2 (x xr ): (r Rt )xr 2rRt yr (20)

= V

(22)

Since the minimal distance between any 2D coordinates [x0 , y0 ] to a line Ax + By + C = 0 is given by jAx0 + By0 + C j (A2 + B 2 )1=2 , the minimal distance of the target to LVFG lines can be expressed as 2rR x + (r2 Rt2 )yr yr 2 t r2 x (r Rt )xr 2rRt yr r dmin = s 2 2rRt xr + (r2 Rt2 )yr +1 (r2 Rt2 )xr 2rRt yr
2 2 + yr )Rt 2r(xr =p 2 2 2 (xr + yr )(r + Rt2 )2

denotes UAVs speed. Therefore, the distance where V from the two planning cycles (A,B) to the target is q (23) x1 = Rt2 + a2 1 Rt q x2 = Rt2 + a2 (24) 2 Rt : For the linear tangent line, the averaged difference is given by 1 (25) TVFG = 1 2 x1 + 2 x2 : Therefore, the maximal objective bias will be obtained when x1 = x2 , such that q 2 2 R : =1 4Rt2 + V (26) TVFG t 2

r2

2r 2 R: + Rt2 t

(21)

Since tangent lines are available, it implies r > Rt . Therefore, from (21), we have dmin > Rt , and the minimal distance depends on the UAV position [xr , yr ]. If the UAV is far away from the target (i.e., r Rt ), this distance could be twice the radius of the objective circle. APPENDIX II PROOF
OF

Now, we consider the case of LVFG alone. Since clockwise LVFG and counterclockwise LVFG are symmetric, we only show the clockwise LVFG without loss of generality. Assume that a target is located at (0, 0), andq that the LVFG trajectory is given

PROPOSITION 2

First, we show the result for T+LVFG. Since the guidance law when a UAV is within an objective circle is the same for T+LVFG and LVFG alone, we compare T+LVFG and LVFG only when the UAV starts outside the objective circle and attempts to approach and fly along it.
852

by fxi , yi g with ri = xi2 + yi2 ; i = 1, 2, : : :. Based on the clockwise LVFG (2), the (k + 1)th trajectory point can be derived from the k th one, such that _k xk xk+1 x = + _k yk+1 yk y xk xk = Hk yk yk 2 2rk Rt xk 1 (rk Rt2 ) = 2 2 2rk Rt 1 (rk Rt ) yk (27)

IEEE TRANSACTIONS ON AEROSPACE AND ELECTRONIC SYSTEMS VOL. 49, NO. 2 APRIL 2013

where Hi =

ri2 Rt2 2ri Rt

2ri Rt ri2 Rt2

[3]

(28)

The squared distance from UAV to the target can be recursively expressed as
2 rk +1 =

Ferng, C., et al. Distributed simulation of forward reachable set-based control for multiple pursuer UAVs. Proceedings of the SimTecT 2006 Simulation Conference and Exhibition, Melbourne, Australia, May 29June 1, 2006. Ma, G., Duan, H., and Liu, S. Improved ant colony algorithm for global optimal trajectory planning of UAV under complex environment. International Journal of Computer Science and Applications, 4, 3 (2007), 5768. Winstrand, M. Mission planning and control of multiple UAVs. Swedish Defence Research Agency, Scientific Report, 16501942, Oct. 2004. Rathinam, S., et al. An architecture for UAV team control. Proceedings of the 5th IFAC Conference on Intelligent Autonomous Vehicles, Lisbon, Portugal, July 57, 2004. Wise, R. UAV control and guidance for autonomous cooperative tracking of a moving target. Ph.D. research proposal, Dept. of Aeronautics and Astronautics, University of Washington, Seattle, WA, June 2006. Cowling, I., Whidborne, J., and Cooke, A. Optimal trajectory planning and LQR control for a quadrotor UAV. Proceedings of the International Conference on Control 2006, Glasgow, Scotland, Aug. 30Sept. 11, 2006. Collins, G., et al. Cooperative control of UAVs for tracking moving targets through information gain. Toyon Research Corporation, Phase II STTR Final Report, Dec. 28, 2007. Tang, S. and Conway, B. Optimization of low-thrust interplanetary trajectories using collocation and nonlinear programming. AIAA Journal of Guidance, Control, and Dynamics, 18, 3 (1995), 599604. Geiger, B., et al. Optimal path planning of UAVs using direct collocation with nonlinear programming. AIAA Guidance, Navigation, and Control Conference, Keystone, CO, Aug. 2006, AIAA Paper 2006-6199. Karatas, T. and Bullo, F. Randomized searches and nonlinear programming in trajectory planning. Proceedings of the 40th IEEE Conference on Decision and Control, vol. 5, Orlando, FL, Dec. 47, 2001, pp. 50325037. Betts, J. T. Survey of numerical methods for trajectory optimization. Journal of Guidance, Control, and Dynamics, 21, 2 (1998), 193207. Faiz, N., Agrawal, S. K., and Murray, R. M. Trajectory planning of differential systems with dynamics and inequalities. AIAA Journal of Guidance, Control, and Dynamics, 24 (Apr. 2001), 219227. Ablavsky, V., Stouch, D., and Snorrason, M. Search path optimization for UAVs using stochastic sampling with abstract pattern descriptors. Proceedings of the AIAA Guidance Navigation and Control Conference, Austin, TX, Aug. 2003. 853

[4]

xk+1 yk+1

xk+1 yk+1

(29)
[5]

2 2 2 2 4 Rt2 )]2 rk + 42 = [1 k (rk k Rt rk :

The nominal speed of the UAV can be computed from 2D vectors, namely s _i T x _i x = V _i _i y y s xi T T xi Hi Hi = i yi yi = i (ri2 + Rt2 )ri : (30)

[6]

[7]

The normalization factor can then be determined by V i = : ri (ri2 + Rt2 ) (31)

[8]

With (29) and (31) the squared distance can be further deduced into 2 2 2 2 2 2 2Vrk (rk Rt ) : = r + V (32) rk +1 k 2 + R2 rk t As LVFG converges, k ! 1, the steady-state distance from UAV to the target may be solved as below.
2 2 2R 2 r VR 2r3 Vr t t = 0:

[9]

[10]

(33)
[11]

The real solution in the third-order polynomial equation is given by q q p p 3 3 r = q + q2 + p3 + q q2 + p3 + 1 6 V (34) where p = and q = 1 2 3 VRt . Therefore, the ASB of the LVFG can be = r Rt . obtained as LVFG
REFERENCES [1] Eberhart, R. and Shi, Y. Particle swarm optimization: Developments, applications and resources. Proceedings of the 2001 IEEE Congress on Evolutionary Computation (CEC 2001), vol. 1, Seoul, Korea, May 2730, 2001, pp. 8186. Rasmussen, S. J., et al. Introduction to the MultiUAV2 simulation and its application to cooperative control research. Proceedings of the 2005 American Control Conference, Portland, OR, June 810, 2005, pp. 44954496.
1 2 2 V 36 1 2 3 Rt 1 3 3 V 216

[12]

[13]

[14]

[2]

[15]

CHEN, ET AL.: UAV PATH PLANNING WITH TANGENT-PLUS-LYAPUNOV VECTOR FIELD GUIDANCE

[16]

Ferguson, D. and Stentz, A. The delayed D* algorithm for efficient path replanning. Proceedings of the 2005 IEEE International Conference on Robotics and Automation (ICRA 2005), Barcelona, Spain, Apr. 1822, 2005, pp. 20452050. ORourke, K., et al. Dynamic routing of unmanned aerial vehicles using reactive tabu search. Military Operations Research Journal, 6 1 (2001), 530. Flint, M., Fernandez-Gaucherand, E., and Polycarpou, M. A probabilistic frame-work for passive cooperation among UAVs performing a search. Proceedings of the Sixteenth International Symposium on Mathematical Theory of Networks and Systems, Leuven, Belgium, 2004. Yang, Y., Polycarpou, M., and Minai, A. Decentralized cooperative search by networked UAVs in an uncertain environment. Proceedings of the 2004 American Control Conference, vol. 6, Boston, MA, June 30July 2, 2004, pp. 55585563. Rathbun, D., et al. An evolution based path planning algorithm for autonomous motion of a UAV through uncertain environments. Proceedings of the AIAA 21st Digital Avionics Systems Conference, vol. 2, Irvine, CA, Oct. 2003, pp. 8D218D212. Barlow, G. J., Oh, C. K., and Grant, E. Incremental evolution of autonomous controllers for unmanned aerial vehicles using multi-objective genetic programming. Proceedings of the 2004 IEEE Conference on Cybernetics and Intelligent Systems (CIS), vol. 2, Singapore, Dec. 2004, pp. 689694. Fradkov, A. and Andrievsky, B. UAV guidance system with combined adaptive autopilot. Proceedings of the IASTED International Conference on Intelligent Systems and Control, Salzburg, Austria, 2003, pp. 9193. Roskam, J. Airplane Flight Dynamics and Automatic Flight Controls, Parts I & II. Lawrence, KS: DAR Corporation, 1998. Stentz, A. Map-based strategies for robot navigation in unknown environments. Planning with Incomplete Information for Robot Problems: Papers from the 1996 AAAI Spring Symposium, Palo Alto, CA: AAAI Press, 1996, pp. 10116. Burns, B. and Brock, O. Information theoretic construction of probabilistic roadmaps. Proceedings of the 2003 IEEE/RSJ International Conference on Intelligent Robots and Systems, vol. 1, Las Vegas, NV, Oct. 2731, 2003, pp. 650655. Mclain, T. and Beard, R. Trajectory planning for coordinated rendezvous of unmanned air vehicles. Proceedings of the AIAA Guidance, Navigation, and Control Conference, Denver, CO, Aug. 1417, 2000, pp. 12471254. Campbell, M. and Ousingsawat, J. On-line estimation and path planning for multiple vehicles in an uncertain environment. Proceedings of the AIAA Guidance, Navigation, and Control Conference and Exhibit, Monterey, CA, Aug. 58, 2002, AIAA Paper 2002-4466.

[28]

[17]

[29]

[18]

[30]

[19]

[31]

[20]

[32]

[21]

[33]

[34]

[22]

[35]

[23]

[36]

[24]

[37]

[25]

[38]

[26]

[39]

[27]

[40]

Misovec, K., et al. Low observable nonlinear trajectory generation for unmanned air vehicles. Proceedings of the 42nd IEEE Conference on Decision and Control, Piscataway, NJ, 2003, pp. 31033110. Lee, J., et al. Strategies of path-planning for a UAV to track a ground vehicle. Proceedings of the 2nd Annual Symposium on Autonomous Intelligent Networks and Systems (AINS 2003), Menlo Park, CA, June 30July 1, 2003. Pongpunwattana, A. and Rysdyk, R. Real-time planning for multiple autonomous vehicles in dynamic uncertain environments. Journal of Aerospace Computing, Information, and Communication, 1, 12 (Dec. 2004), 580604. Frew, E. and Lawrence, D. Cooperative stand-off tracking of moving targets by a team of autonomous aircraft. Proceedings of the AIAA Guidance, Navigation, and Control Conference and Exhibit, San Francisco, CA, Aug. 1518, 2005, AIAA Paper 2005-6363. Rysdyk, R., Lum, C., and Vagners, J. Autonomous orbit coordination for two unmanned aerial vehicles. Proceedings of the AIAA Guidance, Navigation, and Control Conference and Exhibit, San Francisco, CA, Aug. 1518, 2005, pp. 48764884. Milam, M. and Mushambi, R. A new computational approach to real-time trajectory generation for constrained mechanical systems. Proceedings of the IEEE Conference on Decision and Control, Sydney, Australia, Dec. 2000, pp. 845851. Singh, L. and Fuller, J. Trajectory generation for a UAV in urban terrain, using nonlinear MPC. Proceedings of the 2001 American Control Conference, vol. 3, Arlington, VA, June 2527, 2001, pp. 23012308. Becker, M., Dantas, C., and Macedo, W. Obstacle avoidance procedure for mobile robots. In ABCM Symposium Series in Mechatronics (1st ed.), vol. 2, Brazilian Society of Mechanical Sciences and Engineering, Rio de Janeiro, 2006, pp. 250257. Keerkov, S. M., Kabamba, P. T., and Zeitz, III, F. H. Optimal path planning for unmanned combat aerial vehicles to defeat radar tracking. Journal of Guidance, Control, and Dynamics, 29, 2 (Apr. 2006), 279288. Richards, A. and How, J. Aircraft trajectory planning with collision avoidance using mixed integer linear programming. Proceedings of the 2002 American Control Conference, vol. 3, Anchorage, AK, May 810, 2002, pp. 19361941. Chen, H., Chang, K. C., and Agate, C. A dynamic path planning algorithm for UAV tracking. Proceedings of SPIE, Signal Processing, Sensor Fusion, and Target Recognition XVIII, vol. 7336, Orlando, FL, Apr. 2009. Chen, H., Chang, K. C., and Agate, C. Tracking with UAV using tangent-plus-Lyapunov vector field guidance. Proceedings of the 12th International Conference on Information Fusion, Seattle, WA, July 69, 2009, pp. 363372. Lawrence, D. Lyapunov fields for UAV track coordination. Proceedings of the 2nd AIAA Un-manned Unlimited Systems, Technologies and Operations Conference, San Diego, CA, Sept. 2003.

854

IEEE TRANSACTIONS ON AEROSPACE AND ELECTRONIC SYSTEMS VOL. 49, NO. 2 APRIL 2013

[41]

[42]

[43]

[44]

Frew, E., Lawrence, D., and Morris, S. Coordinated standoff tracking of moving targets using Lyapunov guidance vector fields. Journal of Guidance, Control, and Dynamics, 31, 2 (Mar.Apr. 2008). McGee, T., Spry, S., and Hedrick, J. Optimal path planning for an aircraft with a bounded turn rate in the presence of a constant wind. Proceedings of the AIAA Conference on Guidance, Navigation, and Control, Aug. 2005. Nikolos, I. K., et al. Evolutionary algorithm based offline/online path planner for UAV navigation. IEEE Transactions on Systems, Man, and Cybernetics, Part B: Cybernetics, 33, 6 (Dec. 2003), 898912. Petillot, Y., Ruiz, I. T., and Lane, D. M. Underwater vehicle obstacle avoidance and path planning using a multi-beam forward looking sonar. IEEE Journal of Oceanic Engineering, 26, 2 (Apr. 2001), 240251.

[45]

[46]

[47]

[48]

Lorusso, A. and Micheli, E. D. An approach to obstacle detection and steering control from optical flow. Proceedings of the 1996 IEEE Intelligent Vehicle Symposium, Tokyo, Sept. 1920, 1996, pp. 357362. Gandhi, T., et al. Detection of obstacles in the flight path of an aircraft. IEEE Transactions on Aerospace and Electronic Systems, 39, 1 (Jan. 2003), 179191. Kavraki, L., et al. Probabilistic roadmaps for path planning in high-dimensional configuration spaces. IEEE Transactions on Robotics and Automation, 12, 4 (1996), 556580. Larson, J., et al. Advances in autonomous obstacle avoidance for unmanned surface vehicles. AUVSI Unmanned Systems North America 2007, Washington, D.C., Aug. 79, 2007.

CHEN, ET AL.: UAV PATH PLANNING WITH TANGENT-PLUS-LYAPUNOV VECTOR FIELD GUIDANCE

855

Hongda Chen received his B.S. and M.S. degrees in electrical engineering from the Beijing Institute of Technology, Beijing, China in 1987 and 1990, respectively, and his Ph.D. degree in information technology from George Mason University, Fairfax, VA in 1999. From 1998 to 2002 he was a senior member of technical staff in Mobile Satellite Communication, Hughes Network Systems. In 2003 he joined Intelligent Automation as a senior research engineer. He performed state-of-the-art research and development in Riorey, Inc. for advanced network security products in 2006. Since 2003 he has served as a senior research consultant on projects of multisensor data fusion and multitarget tracking in the Department of SEOR, George Mason University, VA, and currently, he is a senior research scientist with Sigma Space Corporation, Lanham, MD, where he works on NASA EOS projects. His research interests include signal processing, data analysis, communications, and fault diagnosis/prognosis in industry applications. Dr. Chen has published more than 30 papers in the areas of signal processing, data fusion, and communications.

Kuo-Chu Chang (M86SM95F10) received his M.S. and Ph.D. degrees in electrical engineering from the University of Connecticut, Storrs, in 1983 and 1986,respectively. From 1983 to 1992 he was a senior research scientist in the Advanced Decision Systems (ADS) division, Booz-Allen & Hamilton, Mountain View, CA. In 1992 he joined the Department of Systems Engineering and Operations Research, George Mason University, where he is currently a professor. His research interests include estimation theory, optimization, signal processing, and multisensor data fusion. He is particularly interested in applying unconventional techniques in the conventional decision and control systems. Dr. Chang has more than 30 years of industrial and academic experience and has published more than one hundred and sixty papers in the areas of multitarget tracking, distributed sensor fusion, and Bayesian Networks technologies. He was an associate editor on Tracking/Navigation Systems from 1993 to 1996 and on Large Scale Systems from 1996 to 2006 for IEEE Transactions on Aerospace and Electronic Systems. He was also an Associate Editor of IEEE Transactions on Systems, Man, and Cybernetics, Part A from 2002 to 2007. He is a member of Etta Kappa Nu and Tau Beta Pi.

Craig Agate received his B.S. and M.S. degrees in electrical engineering from California State University, Northridge and his Ph.D. degree in electrical engineering from the University of California, Santa Barbara in 2000. Upon graduating he joined Toyon Research Corporation in Goleta, CA. With a general background in detection and Bayesian estimation theory, he has developed a particle filtering algorithm to address the unique aspects of ground target tracking (incorporating road constraints, evidence such as nondetection events, and measured signatures for feature-aided tracking and identification). Other research includes developing feature-aided tracking algorithms and object identification algorithms based on Bayesian evidence accrual. Addressing the need for long-term continuous tracking of particular targets, he has also developed an approach to feature-aided tracking that exploits features collected on-the-fly to characterize targets of interest, which has provided a fingerprint that allows for disambiguation of the target of interest from background targets.
856 IEEE TRANSACTIONS ON AEROSPACE AND ELECTRONIC SYSTEMS VOL. 49, NO. 2 APRIL 2013