VOL. 10, NO. 4

Direct Trajectory Optimization Using Nonlinear Programming and Collocation
C.R. Margraves* and S.W. Paris* Boeing Aerospace Company, Seattle, Washington

An algorithm for the direct numerical solution of an optimal control problem is given. The method employs cubic polynomials to represent state variables, linearly interpolates control variables, and uses collocation to satisfy the differential equations. This representation transforms the optimal control problem to a mathematical programming problem which is solved by sequential quadratic programming. The method is easy to program for a very general trajectory optimization problem and is shown to be very efficient for several sample problems. Results are compared with solutions obtained with other methods.

Introduction DIRECT trajectory optimization method that represents te and control variables by piecewise polynomials is described in this paper. An implicit integration scheme based on Hermite interpolation is used to convert the optimal control problem to a nonlinear programming problem. The primary advantage of this method is that it is much easier to extend to general problems involving path constraints, discontinuous states, and control inequalities. Numerous approaches are possible for direct trajectory optimization. The authors have described in earlier papers a method employing piecewise Chebyshev polynomials. Penalty functions were used to convert optimal control problems to unconstrained parameter optimization problems which were solved with a full second-order modified Newton algorithm.1'3 The basic ideas used in this approach were described earlier by Johnson4 and Hahn and Johnson.5 A similar approach was described by Balakrishnan.6 Additional direct approaches were described by Kelley,7 Rader and Hull,8 and Brauer et al.9 Birkbff and de Boor10 hae shown that piecewise polynomials have many desirable properties for representing smooth curves, de Boor11 and Russell and Shampine12 described implicit methods for the solution of boundary value problems in ordinary differential equations using piecewise polynomials. Dickmanns and Wells13 and Dickmanns14'15 applied the maximum principal to reduce the optimal control problem to a boundary value problem which they solved using a collocation method; the method employed Hermite interpolation and collocation.16 The solution of optimal control problems using nonlinear programming has been demonstrated for a number of distinctive methods. Betts et al.17 described a procedure in which a problem is divided into a series of subproblems which are solved by an indirect method. Boundary conditions on the subproblems are solved by nonlinear programming. Kraft18 parameterized the control variables and integrated explicitly the equations. A direct and an indirect method were utilized and compared. Evtushenko19 used nonlinear programming to define a variety of optimal control algorithms. Renes20 proposed a procedure which is similar to that described in the next section. Renes did not give any numerical results and a

number of details are different (e.g., different order splines are used, the collocation points are different, the independent variables were spline coefficients, etc.).

Mathematical Method
Statement of the Problem

Trajectory optimization problems that can be described by a sequence of vehicle/flight stages are considered. The control functions u(t), the time points delimiting the various stages (Eit called events), and the vehicle design parameters (co) are sought which minimize a given performance index J=<t>[x(E)tu(E),u,E]

The /th stage is assumed to be a dynamical system subject to the differential constraint ' =/ (x,u,u,t)

on the interval te[EifEi+l]t where * is a vector of states governed by first-order differential equations, u a vector of controls (e.g., pitch angle, and o> a vector of design parameters (e.g., planform area, rocket nozzle diameter, etc.). The prime denotes differentiation with respect to time. At each event Eit nonlinear boundary conditions may be imposed of the form

where £BE is the vector lower-limit on the boundary conditions and (7BE is the upper-limit. Equality constraints are possible by setting the appropriate elements of £BE and £/BE equal to each other. Path constraints of the form

may also be imposed on the system. Denote the collection of events by ET= (El9E2,...9EN+l), where N is the number of stages. The functions/, a', and hk are assumed to be smooth (class C2 or better) within a given stage. However, no continuity restrictions are imposed at events. Specifically, discontinuities of the form

Received June 9, 1986, revision received Dec. 10, 1986. Copyright © American Institute of Aeronautics and Astronautics, Inc., 1987. All rights reserved. * Senior Specialist Engineers. Members AIAA.

are allowed at the events. The 6,'s may be fixed quantities or may be included in the design parameter set o>. Equations (5)

as defined by the equations. the functions x(t) and u ( t ) are represented by piecewise polynomials and collocation is used to satisfy Eq.1 00 Q C2 _C3. quintic polynomials for the solution of secondorder differential equations (without reduction to the first order) were tried.JULY-AUGUST 1987 DIRECT TRAJECTORY OPTIMIZATION 339 account for physical events such as weight jettisons and impulsive velocity changes. The basic procedure can be derived as follows. (7) 1111 0 1 2 3 It is now noted that our payoff function / is a function of P only.22 and Reklaitis. (3)] and the constraints [Eq. (4) where to simplify the argument. (6) at S= 1/2. (6) and evaluating at 0 and 1 gives " 1 0 0 0 " " Co " 0. collect all of the independent variables into a single vector P defined by PT=[ZT. Linear interpolation is used to obtain the controls at the segment centers. cubic polynomials are defined for each state on each segment using values of the states at the nodes (the boundaries at the segments) and the state time derivatives. Recent treatments of the subject are presented by Gill et al. SQP methods were popularized by Biggs. connecting the segments constitute additional linear equations to be satisfied. at the nodes. Method of Solution The approach is to reduce the optimal control problem stated in the previous section to a nonlinear programming problem. NPDOT uses a nonlinear programming package called NPSOL which was developed by the Systems Optimization Laboratory at Stanford University. Let the states x be represented on each segment by cubics of the form The defects for each state evaluated at the center of each segment constitute a set of nonlinear algebraic equations which are a function of the states and controls at each node.-*.21 Evtushenko. Equations (5). AP is composed of all the linear relationships from Eqs.) In the same way.25 Han. the interval [EifEi+ {] is subdivided into Ns segments/Let the ratio of the length of theyth segment to T. and transforming to segment length T=Tjit Tsi. The trajectory optimization problem stated above can be expressed as Inverting the 4 x 4 matrix gives "QQ C2 C3 " 1 0 0 1 0 0 0 " 0 (8) subject to (14) -3 -2 2 3 -1 1 1 -2 Now using Eq. these derivatives are computed by finite difference. (2) at xc gives fc. In the method described here. (2). (3) HN = collection of all nonlinear relationships from Eq. This procedure is illustrated in Fig. Equation (14) constitutes a nonlinear programming problem.23 These books contain long lists of references on earlier work in this field. Also. Using Hermite interpolation. The length of each state is defined as Tsi=Ei+l -Ei9 as shown in Fig. Subroutines must be provided which evaluate <t>(P)t C(P). This approach did not work as well as the cubics. = oo. For equality constraints £/ = «/. Evaluating Eq.26 and Powell. we see that the interpolated value of x at the center of the segment is (9) where// is Eq. dx/ds(Q) = xl9 dx/ds(l) = x2. If the cubic polynomial is capable of representing the solution on the given segment. The boundary conditions [Eq.ET. For each stage. Differentiating Eq. This task is greatly simplified by the fact that the defects depend on states at adjacent nodes only and only . (3-5). and their first derivatives with respect to P. evaluating Eq. A large amount of literature in recent years has been devoted to the solution of nonlinear programming problems. the segment length S is transformed such that Se[0.27 NPSOL requires as inputs the matrix A and an initial guess for P. the length of the yth segment in the fth stage is r^T^. (2). 1].24 NPSOL uses sequential quadratic programming (SQP). the events E9 and the design parameters co. Define the defect at the center of the segment as A =/. (2) evaluated at xt (the superscript denoting the stage was dropped for simplicity). (8). 2. Now. The values of the states are then selected (by nonlinear programming) to force the interpolated derivatives to agree with the differential equations at the center of the segment. then selecting x1 and x2 to drive A to zero will produce an accurate approximation to the solution ofEq. The computer code implementing the aforementioned procedure is called Nonlinear Programming for Direct Optimization of Trajectories (NPDOT). and inequality constraints are handled by £/= — oo or «. (11) x{ and x2 are varied to drive A = 0.uT] (12) where x = C0 + Q S + C2S + C3S 2 3 (6) Collecting all of the nonlinear equations into a single vector equation gives (13) where A/:/ = defect for rth state at yth node BN = collection of all nonlinear relationships from Eq. 1. Programming Considerations -x2)/2T- (10) is obtained. £and u are the upper and lower bounds. u(tt). (4)] evaluated at both the nodes and centers of the segments provide additional equations (constraints) to be satisfied. the slope where NP is the dimension of P. Thus.. (Note x^ = Tf[xt. To do this. Initial experiments with cubic polynomial controls did not work well.^ be denoted as Tyi. Let x(Q) = xl9 x(l) = x2. tif a?].

To obtain accurate solutions to realistic problems. PARIS J. 2) Brachistochrone (Ref. 341-345). . The Jacobian has the banded structure shown in Fig. data smoothing. 3) minimum time interceptor (Ref. An initial node selection can be made based on the nominal. to gain experience before attempting new unknown solutions. Regions where large changes occur as a function of Mach number for the nominal and optimal solution may be very different regions in time. The key refinements involve problem scaling. 3. and node refinement. 33). " ^ ^ ^ ^ =m ^ J * CTOp multiple runs • NPDOT single runs • All times are Cray X-MP/24 CPU seconds • CTOP . This can be difficult to accomplish since the nodes are distributed in thie whereas tabular data may be functions of other variables such as Mach number of altitude. and to demonstrate the method's inherent flexibility. After several refinement cycles.Program to Optimize Simulated Trajectories (Martin) Fig. The times shown in Fig.340 C. Partial derivatives must be accurate to obtain convergence of the nonlinear programming algorithm. 1. The code described here uses this algorithm initially and retains the perturbation step sizes for subsequent iterations. Unless a nominal that is close to opitimal is known. The smoothness of data is closely related to node spacing since. some refinements must be added to the procedure. The following scaling procedure worked well. The basic idea is to represent each segment by a higher-order method (polynomial. 2 Implicit integration illustration.8 Constrained Brachistochrone CTOP NPDOT CTOP NPDOT • CTOP. 34). 4. It is usually preferable to compute derivatives directly by finite difference. The Chebyshev Trajectory Optimization Program (CTOP). 31 and 32). 4 CPU time comparison. etc. The node selection can then be updated after initial convergence or when optimizer behavior indicates possible problems due to the polynomial distribution. 28. A discussion of equation scaling as applied to optimization problems may be found in Ref. Numerical Examples Several test problems that have been solved with the new technique are described in this section. (For approximate comparison with other computers. A comparison of the solution times between NPDOT and CTOP is given in Fig. A stepped approach was used to validate NPDOT against known results. it is important that they be smooth. in general. NPDOT min time to climb = . Fig.) States and controls were scaled by The following problems have been solved using NPDOT: 1) Van der Pohl (Ref.) and add nodes when disagreement is found. B.R. an adaptive procedure is requried. the independent variables should be scaled so that they have similar mangitudes. MARGRAVES AND S. served as the performance reference and "truth" model for NPDOT. • POST . 3 Jacobian structure.) Discussions of prob- where XN is the nominal trajectory and Sx is an estimate of an upper bound for X-XN. Fig. Smoothing data without changing the physical meaning can be a difficult problem. Good results were obtained with splines [see Ref.C. partial derivative computation. For realistic trajectory problems. rapid changes in tabular functions will be acceptable only if there are sufficient nodes in the region of rapid change. node selection. the adequacy of the total representation can be treated by explicit numerical integration of the trajectory using the polynomial control function and comparing the results with the polynomial representation of the trajectory. Nodes Events 1 EI 2 3 4 E2 5 10 11 12 Fig. The problems range from simple analytical problems to contemporary aerospace performance-prediction problems. An algorithm for selecting the finite-difference perturbation size is given in Ref. Times were scaled by TS = T/ST where ST is the estimated upper bound for T. Dickmanns14 has described several methods for refining node selection. 28 (pp. 4 are adjusted to be comparable for the Cray XMP/24.Chebyshev Trajectory Optimization Program (Boeing. the X-MP has a clock cycle time of 9 ns. The academic problems were also compared to solutions using the maximum principle. 29)] for tables with one independent variable. 1 Piecewise polynomial representation.30 1. GUIDANCE on the time length of the stage in which it occurs. If tabular data are used. 4) advanced booster. the computation of analytical derivatives can be a formidable task. A satisfactory solution for multivariable tables has not yet been found.W. The results of NPDOT agreed with the previously published results for problems 1-3. (No systematic comparison with other approaches was attempted. numerical integration. described in Ref. As a minimum.

which had a time to climb of 325. 341 2. The thrust was a function of Mach number and altitude. The specific impulse is a function of Mach number. The strategy employed was to accelerate the vehicle to a dynamic pressure <7.18 rad. which matched our previously obtained CTOP result and published energy state results. shown by long dashes. The specific fuel consumption was constant. To counter this activity.500 5.000 7. The thrust and specific impulse were modeled as tabular functions of Mach number. The equations of motion were for three-dimensional flight above a spherical nonrotating Earth. 100 200 Time.000 7. Two additional nodes were added to yield a 15-unequal-segment distribution. to an altitude of 20 km. and angle of attack. CA and CN = CNa -a. the control (pitch) was set to 0. The final propulsion phase. 5. finishing up by a zoom climb to the final altitude.3 s and produced a trajectory.35 The control function and velocity profile are shown in Fig.500 -5.38. The propulsion system has several discrete phases which depend on the vehicle's Mach number. 5 Supersonic interceptor minimum-time climb trajectories. CA and CNa were functions of Mach number. A solution was generated for this 13-segment distribution and another error estimate was produced. T. In the initial attempts for a solution.500 5. This provided the trajectory. One can observe that this trajectory exploits the discretization to meet the constraints at the nodes and centers while violating such things as the altitude constraint in between. shown by short dashes. The values for the defining aerodynamic and propulsion data were taken from Ref. nmi 5. This distribution yielded a time to climb of 317. limit and fly that limit until a predetermined velocity was reached. To obtain the node distribution. Initial attempts to generate trajectories using a conventional explicit integrating program for this class of vehicle required intensive "man-in-the-loop" interactions. The aerodynamic coefficients were functions of Mach number. A sequence of solutions is shown on Fig.500 Chebyshev 8 segment distribution 2. the absolute value of the difference between the time derivatives produced by the equations of mo- This problem involves maximizing the final weight that can be injected into low Earth orbit by an advanced booster which utilizes airbreathing propulsion. was obtained with 16 equal segments.000 Great circle range. lems 3 and 4 follow to provide some insight into the method as applied to typical aerospace flight mechanics problems.000 * 70. A preliminary solution. This is consistent with previous trajectory optimization experience. 5. The program could not reduce the defects or obtain a small gradient.300 > 1. sec 300 Fig. The initial estimate of the free variables was as follows: final range equal to 360. 35.0. 6. then climbs at a constant Mach number followed by a dive through Mach one and beyond. an eight-segment solution with a Chebyshev node distribution was generated.000 20 30 40 50 60 Range.36 Advanced Booster This problem involves finding the pitch function 0(0 to take a supersonic interceptor from sea level.000 Ibs. Mach= 1. The initial weight was 42. 7 Advanced booster trajectory. nmi 7.100 ~ 900 700 > 500 300. a rocket burn is performed which . The second propulsion phase being proportional to dynamic pressure tended to drive the trejectory solutions to very high dynamic pressure values.700 o 1.000 ft and the final mass set to 1204 slugs. At that time.000 30. 50 40 I 30 5 20 | 10 Fig. This did not work well. The second propulsion phase goes from Mach 2-20. Five nodes were added to the center of the segments which were producing the majority of the error. 2. was a rocket in which both the thrust and specific impulse were constant.000. These problems were resolved by using the taut spline and tensor spline packages as defined and coded by de Boor.500 J 1. Supesonic Interceptor Minimum-Time Climb tion and time deriatives of the polynomials were integrated to acquire an error estimate.500 § 1. Limits were imposed on dynamic pressure.000 1110. The thrust coefficient was a function of Mach number and angle of attack. The initial-state values were obtained by linear interpolation between the initial and final values.000 ^150. altitude. This was not as good as the previous CTOP1 results. X. 100 200 300 I" o -20. shown as a solid line in Fig. This problem was proposed and solved by Bryson. 6 Supersonic interceptor control function and velocity profile. The airplane accelerates to a high subsonic speed. in minimum time. The first phase spans from Mach 0-2. Mach = 0. The atmospheric density and speed of sound were taken from the N ACA-1962 atmosphere model. The aerodynamics for this vehicle were done in body axis form.34 The equations of motion were those for planar flight above a flat Earth. This was done over the eightsegment trajectory.2 s. simple piecewise linear and piecewise cubic interpolation for the aero/propulsion data were used. so a solution was obtained with 15 segments.500 70 Fig. Mach numbers greater than 20. The thrust was modeled as a thrust coefficient multiplied by dynamic pressure.JULY-AUGUST 1987 DIRECT TRAJECTORY OPTIMIZATION „ 190.

1972. Nov. A coverged NPDOT trajectory was obtained using 22 segments and resulted in a 10% gain in weight over the baseline trajectory. Conclusions A trajectory optimization method which uses an embedded collocation scheme in conjunction with mathematical programming has been described. SOL 84-7." Approximation of Functions. "User's Guide for NPSOL (Version 2. 7. Numerical Optimization Techniques. 81.E. 1980. 6. 1978. Academic Press. and Zondervan. Bjdrck. Numerical Optimization Techniques. Cornick. Practical Optimization. edited by G. A. Hoffman. "Application of an Advanced Trajectory Optimization Method to Ramjet Propelled Missiles. of Cambridge. and Hoffman.H. Academic Press. Part 1: General Formulation. C. the Netherlands. Technical Rept.." Mathematical Programming. R.C. S. 1984. 17 Betts..." Hochschule der Bundeswehr Munchen. 1983. July-Aug..C. "Collocated Hermite Approximation of Third and Fifth Order with Automatic Mesh Refinement for the Solution of Optimal Control Boundary Value Problems.G.T. Seattle.E." 2nd IFAC Workshop on Optimization.R. W. F." The Boeing Co. 5 Hahn.. Prentice-Hall Englewood Cliffs. 1985. 411-428. When NPDOT was applied to this problem. Div. and Wright.D. and Denham. D.. Vol. and Wells. New York. K. "Numerical Computation of Optimal Atmospheric Trajectories Involving Staged Vehicles. 1978. Stanford Univ. Acknowledgment This research was supported in part by the U. Numerical Methods. the vehicle's final conditions are orbital. 1980..C.H. Vol. M. 18 Kraft. W. MA. "Numerical Computation of Optimal Atmospheric Trajectories. "Adaptations of the Conjugate Gradient Method to Optimal Problems with Terminal State Constraints. 149-173... New York. Iowa State Univ. "Efficient Convergence and Mesh Refinement Strategies for Solving General Ordinary Two-Point Boundary Value Problems by Collocated Hermite Approximation. 341-345. 11. 25 Biggs. K.. H. 1970. and Rettie. J. 196-364. Blaisdell Publishing. 1978.. England. "Computation of Optimal Aircraft Trajectories Using Parameter Optimization Methods.. 1981. pp. 36 de Boor.R. T... IFIP-TC7. Div. K. 6.F. Vol. Y.R. 299-314.. Interscience. "A Collocation Method for Boundary Value Problems. "On the Choice of Minimization Algorithms in Parametric Optimal Control.V. Vol. pitch profile. 346-356. 285. Feb. Amsterdam.342 C. pp. Johnson. D.. FRG. A. This results in the problem of maximizing the weight injected into a predefined transfer orbit. DAMTP 77/NA 2. the vehicle wants to fly at maximum q\ however. 205-254. 1984.. Murray. Y. Vol. Leitmann. Engineering Optimization.. 35 Bryson.J. Feb. To maximize acceleration. 993-997. Academic Press. No. of Operations Research. 4 Johnson. and drag losses are largest at maximum q so there is a balance between the two. G.R. Springer-Verlag. "A Fast Algorithm for Nonlinearly Constrained Optimization Calculations. Dept. 1975. 7. Stanford. "A Second-Order Feedback Method for Optimal Control Computations.. New York. 27 Powell. "Final Report for Chebyshev Trajectory Optimization Program (CHEBYTOP II). 16 Dahlquist..V.. 15 Dickmanns. New York. London.E. and Wright.. T. and Ho. Gamkrelidze. Gill. 1969.S. Rept. pp. Springer-Verlag. A Practical Guide to Splines. * 'Approximate Finite-Thrust Trajectory Optimization. "Constrained Minimization Using Recursive Equalty Quadratic Programming.. 26 Han. and Anderson. 1976. A Practical Guide to Splines. Academic Press. edited by F.L...L.. Jan. GUIDANCE allows the vehicle to coast in a transfer orbit with the appropriate apogee. 1977. Dec. NJ..D.. 1. Pub. 406-414. 12 Russell. 4. C. 6.. New York. D. P. Sept. 29.E. Vol. and Johnson.W. "Solving the Optimal Control Problem Using a Nonlinear Programming Technique. HARGRAVES AND S. Oct.. CA. pp. B. 1975.W.. 3 Hargraves.. 2 Paris.. P.. 1974.E.D. Murray.. 1981. Bauer. E. pp.G.. New York. Optimization Software. R. "On the Use of Splines and Collocation in a Trajectory Optimization Algorithm Based on Mathematical Programming. Joosten. Paris. C. WS. 10 Birkhoff. This was done to benchmark NPDOT with the existing solution. Springer-Verlag. C." Journal of Aircraft. and Hull... 1981...R. 9 Brauer. 19. L. C. 13 Dickmanns. Murray. J.R. M.J. M. only the initial airbreathing portions and rocket burn were optimized.A.-Dec.. E. Technical Rept. June 1971. AIAA 20th Aerospace Sciences Meeting. 33 Bryson. L..E.E. 23 Reklaitis.. J. Desai. Vol. C. AC-12. 31 Bullock.E. G." IEEE Transactions on Automatic Control. S. pp..." Numerical Mathematics. New York. W. G." SIAM Journal on Control. Oct." AIAA Journal. The Mathematical Theory of Optimal Processes. Elsevier.. Nov. 21 Gill. G." Journal of Applied Mechanics. A. pp. F. and de Boor. New York. Inc." Systems Optimization Lab. FL. Jan. pp. and Wright. 864-866." AIAA Paper 82-0360.F.. V. pp. The best results were achieved by using standard trade studies... 1965..H. 1985. pp. 30 Pontryagin. A. Saunders. 1982.A. G. "Capabilities and Applications of the Program to Optimize Simulated Trajectories (POST). "On a New Computing Method in Optimal Control.K. 1-28. Oberpfaff-Enhofen." Journal of Aircraft.. New York. Garabedian.E. 28 . Aug. Optimization Software. PARIS J." Optimization Techniques.V..K." Numerical Methods for Nonlinear Optimization. 22 Evtushenko. The method has been used to solve a wide variety of test cases and found to be superior to other procedures in terms of cost and robustness... Ravindran. "Superlinearly Convergent Variable Metric Algorithms for General Nonlinear Programming Problems. p. Attempts to use CTOP on this problem resulted in very slowly converging trajectories.J. edited by H. and Fink.. 1977.. 119-120.-Dec." Engineering Research Institute. M." Optimal Control Applications and Methods. 481-488. All attempts to optimize this trajectory with an explicit integrating parameter-optimization program failed. "A Steepest-Ascent Method for Solving Optimum Programming Problems. Practical Optimization. and Stevenson. E.T. Springer-Verlag. 1981. S.R. p. New York. LRT/WE. A. prepared for NASA under Contract NAS2-5994. The trajectory. M.. Springer-Verlag. It is interesting to note that the optimum trajectory does not stay on the dynamic pressure constraint but rather initially rides it and then gets off.M. References Margraves.." Univ. D. 20 Renes. 1969. D180-12916-1. Pub. June 1962. NLRTR-78016 U. Vol.. 29 de Boor. 34 Bryson.. Air Force under Contract F33615-85-C-3009. 319-334. 1967. M. n de Boor. W. 24 Gill. W. 388-398. 8 Rader J. Inc. Vol. pp.. I. P. 6 Balakrishnan..D. "Energy-State Approximation in Performance Optimization of Supersonic Aircraft. and dynamic pressure as a function of time are shown in Fig.. 12. June 1969. L.H. F. 7 Kelley... New York. Wiley. Waltham. A single apogee burn is then used to circularize the orbit. 263-282.F. 1972.E. pp. May 1968.." AIAA Paper 84-2037. pp.. R.T. pp. Y. Boltyanskii. 1979. The transfer orbit is defined by a fixed velocity and flight path angle at a given altitude.G.. 32 Willoughby. FRG. A. 19 Evtushenko. "Piecewise Polynomial Interpolation and Approximation... 1962. 1962. 1981.S." NASA CR-2770." National Aerospace Lab. and Mishchenko. pp. F.. M.W. Rept." AIAA Journal of Guidance and Control. W. N. Vol.P.. A Practical Guide to Splines. Sept.F. Lootsma.W." Proceedings of the 6th Technical Conference on Optimization Techniques. "Method of Gradients. Sept.D. ERI-62500. Applied Optimal Control. and Ragsdell.. 14 Dickmanns. New York. "Approximate Solution of Optimal Control Problems Using Third Order Hermite Polynomial Functions..1): A FORTRAN Package for Nonlinear Programming. and Shampine. Orlando.N. and Franklin. New York.

Sign up to vote on this title
UsefulNot useful