Complete Example: Time-Optimal Bilinear Control

This example demonstrates solving a time-optimal trajectory optimization problem with:

  • Multiple control inputs with bounds
  • Free time steps (variable Δt)
  • Combined objective (control effort + minimum time)
using DirectTrajOpt
using NamedTrajectories
using LinearAlgebra
using CairoMakie

Problem Setup

System: 3D oscillator with 2 control inputs

\[\dot{x} = (G_0 + u_1 G_1 + u_2 G_2) x\]

Goal: Drive from [1, 0, 0] to [0, 0, 1] minimizing ∫ ||u||² dt + w·T

Constraints:-1 ≤ u ≤ 1, 0.05 ≤ Δt ≤ 0.3

Define System Dynamics

G_drift = [
    0.0 1.0 0.0;
    -1.0 0.0 0.0;
    0.0 0.0 -0.1
]

G_drives = [
    [
        1.0 0.0 0.0;
        0.0 0.0 0.0;
        0.0 0.0 0.0
    ],
    [
        0.0 0.0 0.0;
        0.0 0.0 1.0;
        0.0 1.0 0.0
    ],
]

G = u -> G_drift + sum(u .* G_drives)
#2 (generic function with 1 method)

Create Trajectory

N = 50
x_init = [1.0, 0.0, 0.0]
x_goal = [0.0, 0.0, 1.0]
x_guess = hcat([x_init + (x_goal - x_init) * (k/(N-1)) for k = 0:(N-1)]...)

traj = NamedTrajectory(
    (x = x_guess, u = 0.1 * randn(2, N), Δt = fill(0.15, N));
    timestep = :Δt,
    controls = (:u, :Δt),
    initial = (x = x_init,),
    final = (x = x_goal,),
    bounds = (u = 1.0, Δt = (0.05, 0.3)),
)
N = 50, (x = 1:3, u = 4:5, → Δt = 6:6)

Build and Solve Problem

integrator = BilinearIntegrator(G, :x, :u, traj)

obj = (QuadraticRegularizer(:u, traj, 1.0) + 0.5 * MinimumTimeObjective(traj, 1.0))

prob = DirectTrajOptProblem(traj, obj, integrator)

prob
DirectTrajOptProblem
  Trajectory
    Timesteps: 50
    Duration:  7.35
    Knot dim:  6
    Variables: x (3), u (2), Δt (1)
    Controls:  u, Δt
  Objective (2 terms)
         1.0 * QuadraticRegularizer on :u (R = [1.0, 1.0], all)
         0.5 * MinimumTimeObjective (D = 1.0)
  Dynamics (1 integrators)
    BilinearIntegrator: :x = exp(Δt G(:u)) :x  (dim = 3)
  Constraints (4 total: 2 equality, 2 bounds)
    EqualityConstraint: "initial value of x"
    EqualityConstraint: "final value of x"
    BoundsConstraint: "bounds on u"
    BoundsConstraint: "bounds on Δt"
solve!(prob; max_iter = 50)
    initializing optimizer...
      building evaluator: 1 integrators, 0 nonlinear constraints
      dynamics constraints: 147, nonlinear constraints: 0
        integrator 1 jacobian structure: 0.369s
      jacobian structure: 1764 nonzeros (0.391s)
        integrator 1 hessian structure: 0.018s
        computing objective hessian structure (CompositeObjective)...
          sub-objective 1 (QuadraticRegularizer): 0.363s
          sub-objective 2 (MinimumTimeObjective): 0.006s
        objective hessian structure: 0.423s
      hessian structure: 2814 nonzeros (0.441s)
      linear index maps built (0.004s)
      evaluator ready (total: 0.974s)
    evaluator created (1.531s)
    NL constraint bounds extracted (0.024s)
    NLP block data built (0.0s)
    Ipopt optimizer configured (0.009s)
    variables set (0.136s)
        applying constraint: initial value of x
        applying constraint: final value of x
        applying constraint: bounds on u
        applying constraint: bounds on Δt
    linear constraints added: 4 (0.477s)
    optimizer initialization complete (total: 2.177s)

******************************************************************************
This program contains Ipopt, a library for large-scale nonlinear optimization.
 Ipopt is released as open source code under the Eclipse Public License (EPL).
         For more information visit https://github.com/coin-or/Ipopt
******************************************************************************

This is Ipopt version 3.14.19, running with linear solver MUMPS 5.8.2.

Number of nonzeros in equality constraint Jacobian...:     1746
Number of nonzeros in inequality constraint Jacobian.:        0
Number of nonzeros in Lagrangian Hessian.............:     2748

Total number of variables............................:      294
                     variables with only lower bounds:        0
                variables with lower and upper bounds:      150
                     variables with only upper bounds:        0
Total number of equality constraints.................:      147
Total number of inequality constraints...............:        0
        inequality constraints with only lower bounds:        0
   inequality constraints with lower and upper bounds:        0
        inequality constraints with only upper bounds:        0

iter    objective    inf_pr   inf_du lg(mu)  ||d||  lg(rg) alpha_du alpha_pr  ls
   0  3.6847125e+00 1.48e-01 9.15e-02   0.0 0.00e+00    -  0.00e+00 0.00e+00   0
   1  2.6575048e+00 1.30e-01 5.35e-01  -4.0 1.74e+00    -  1.55e-01 1.22e-01f  1
   2  2.2777749e+00 1.17e-01 3.84e+00  -0.6 2.94e+00    -  2.94e-01 1.02e-01h  1
   3  2.0955442e+00 1.06e-01 5.53e+00  -4.0 4.05e+00   0.0 1.52e-01 9.09e-02h  1
   4  2.0592870e+00 1.01e-01 3.34e+01  -0.1 3.42e+00   0.4 4.43e-01 4.88e-02h  1
   5  2.1291118e+00 9.94e-02 2.90e+01   0.7 1.08e+01    -  1.49e-01 3.51e-02f  1
   6  2.1114936e+00 9.07e-02 1.03e+02   0.1 2.63e+00  -0.1 7.35e-01 9.02e-02h  1
   7  2.7041974e+00 7.03e-02 2.93e+02   0.7 4.11e+00    -  8.93e-01 2.31e-01h  1
   8  2.6653828e+00 6.59e-02 2.67e+03   1.5 2.54e+00   2.2 7.76e-01 6.33e-02f  1
   9  2.7240401e+00 8.54e-02 2.69e+03   1.9 7.42e+00   1.7 2.24e-01 1.17e-01f  1
iter    objective    inf_pr   inf_du lg(mu)  ||d||  lg(rg) alpha_du alpha_pr  ls
  10  2.9461081e+00 5.74e-02 1.93e+03   1.4 3.20e+00    -  2.28e-01 2.23e-01f  1
  11  3.2803858e+00 6.38e-02 9.14e+03   2.2 1.35e+00   3.0 6.11e-01 7.64e-01f  1
  12  3.0664818e+00 2.11e-02 5.96e+03   2.1 1.57e-01   4.4 9.81e-01 1.00e+00f  1
  13  3.0614070e+00 2.25e-04 1.75e+03   1.2 2.84e-02   4.8 9.99e-01 1.00e+00f  1
  14  3.0615576e+00 7.64e-08 8.51e+00  -0.4 4.03e-04   4.3 1.00e+00 1.00e+00f  1
  15  3.0615327e+00 4.22e-10 3.14e-01  -2.3 4.61e-05   3.8 1.00e+00 1.00e+00h  1
  16  3.0614530e+00 1.97e-09 1.12e-01  -4.0 4.94e-05   3.4 1.00e+00 1.00e+00h  1
  17  3.0612141e+00 1.77e-08 1.12e-01  -4.0 1.48e-04   2.9 1.00e+00 1.00e+00f  1
  18  3.0604988e+00 1.58e-07 1.12e-01  -4.0 4.44e-04   2.4 1.00e+00 1.00e+00f  1
  19  3.0583666e+00 1.40e-06 1.12e-01  -4.0 1.33e-03   1.9 1.00e+00 1.00e+00f  1
iter    objective    inf_pr   inf_du lg(mu)  ||d||  lg(rg) alpha_du alpha_pr  ls
  20  3.0520892e+00 1.20e-05 1.10e-01  -4.0 3.93e-03   1.4 1.00e+00 1.00e+00f  1
  21  3.0342451e+00 9.23e-05 1.06e-01  -4.0 1.14e-02   1.0 1.00e+00 1.00e+00f  1
  22  2.9878112e+00 5.44e-04 9.66e-02  -4.0 3.09e-02   0.5 1.00e+00 1.00e+00f  1
  23  2.9426176e+00 7.08e-04 8.81e-02  -4.0 7.63e-02   0.0 1.00e+00 4.38e-01h  1
  24  2.9109599e+00 3.99e-04 8.41e-02  -4.0 3.04e-02   0.4 1.00e+00 1.00e+00h  1
  25  2.8927377e+00 4.05e-04 7.96e-02  -4.0 7.03e-02  -0.0 1.00e+00 2.35e-01h  1
  26  2.7550793e+00 4.22e-03 5.57e-02  -4.0 1.60e-01  -0.5 1.00e+00 7.89e-01f  1
  27  2.7157347e+00 9.25e-04 5.47e-02  -4.0 6.67e-02  -0.1 1.00e+00 9.72e-01h  1
  28  2.6714039e+00 1.11e-03 4.99e-02  -4.0 1.58e-01  -0.6 1.00e+00 4.06e-01h  1
  29  2.6142829e+00 5.55e-03 6.17e-02  -3.3 6.56e-01  -1.0 9.07e-01 3.02e-01f  1
iter    objective    inf_pr   inf_du lg(mu)  ||d||  lg(rg) alpha_du alpha_pr  ls
  30  2.5478237e+00 1.43e-03 3.89e-02  -4.0 1.49e-01  -0.6 1.00e+00 7.79e-01h  1
  31  2.5212259e+00 1.36e-03 6.22e-02  -3.2 2.83e-01  -1.1 9.77e-01 2.36e-01h  1
  32  2.4780513e+00 8.37e-04 2.27e-02  -4.0 1.04e-01  -0.7 1.00e+00 9.13e-01h  1
  33  2.4587046e+00 1.21e-03 2.82e-02  -3.2 2.47e-01  -1.1 1.00e+00 3.38e-01h  1
  34  2.3894114e+00 4.59e-02 3.66e-02  -4.0 1.09e+00  -1.6 1.47e-01 5.74e-01h  1
  35  2.3823541e+00 2.78e-02 4.77e-02  -2.9 1.61e+00  -2.1 3.06e-01 4.74e-01f  1
  36  2.3230076e+00 2.74e-03 7.27e-03  -3.3 2.96e-01  -1.7 1.00e+00 9.74e-01h  1
  37  2.3049515e+00 7.52e-03 1.18e-02  -3.5 4.26e-01  -2.1 9.07e-01 7.72e-01h  1
  38  2.2889273e+00 6.30e-03 3.95e-03  -3.7 4.23e-01  -2.6 1.00e+00 7.07e-01h  1
  39  2.2809922e+00 4.56e-03 2.64e-03  -3.8 3.34e-01    -  1.00e+00 7.72e-01h  1
iter    objective    inf_pr   inf_du lg(mu)  ||d||  lg(rg) alpha_du alpha_pr  ls
  40  2.2752159e+00 2.33e-04 3.72e-04  -4.0 7.82e-02    -  1.00e+00 1.00e+00h  1
  41  2.2751951e+00 2.33e-05 2.36e-05  -4.0 2.14e-02    -  1.00e+00 1.00e+00h  1
  42  2.2751890e+00 3.19e-08 1.02e-08  -4.0 7.29e-04    -  1.00e+00 1.00e+00h  1
  43  2.2751890e+00 2.50e-13 1.25e-13  -4.0 2.29e-06    -  1.00e+00 1.00e+00h  1

Number of Iterations....: 43

                                   (scaled)                 (unscaled)
Objective...............:   2.2751889659278057e+00    2.2751889659278057e+00
Dual infeasibility......:   1.2548133105499959e-13    1.2548133105499959e-13
Constraint violation....:   2.5024426975051028e-13    2.5024426975051028e-13
Variable bound violation:   0.0000000000000000e+00    0.0000000000000000e+00
Complementarity.........:   1.0000000000026431e-04    1.0000000000026431e-04
Overall NLP error.......:   2.5024426975051028e-13    2.5024426975051028e-13


Number of objective function evaluations             = 44
Number of objective gradient evaluations             = 44
Number of equality constraint evaluations            = 44
Number of inequality constraint evaluations          = 0
Number of equality constraint Jacobian evaluations   = 44
Number of inequality constraint Jacobian evaluations = 0
Number of Lagrangian Hessian evaluations             = 43
Total seconds in IPOPT                               = 12.178

EXIT: Optimal Solution Found.

Visualize Solution

plot(prob.trajectory) # See NamedTrajectories.jl documentation for plotting options
Example block output

Analyze Solution

x_sol = prob.trajectory.x
u_sol = prob.trajectory.u
Δt_sol = prob.trajectory.Δt

println("Solution found!")
println("  Total time: $(sum(Δt_sol)) seconds")
println("  Δt range: [$(minimum(Δt_sol)), $(maximum(Δt_sol))]")
println("  Max |u₁|: $(maximum(abs.(u_sol[1,:])))")
println("  Max |u₂|: $(maximum(abs.(u_sol[2,:])))")
println("  Final error: $(norm(x_sol[:,end] - x_goal))")
Solution found!
  Total time: 4.2461719205676305 seconds
  Δt range: [0.072857964224019, 0.175]
  Max |u₁|: 0.9967728853584535
  Max |u₂|: 0.9979498454887172
  Final error: 0.0

Key Insights

Free time optimization: Variable Δt allows the optimizer to adjust trajectory speed, with shorter steps where control is needed and longer steps in smooth regions.

Control bounds: With time weight 0.5, controls don't fully saturate. Increase the weight to push toward bang-bang control.

Combined objectives: The + operator makes it easy to balance multiple goals.

Exercises

1. Bang-bang control: Set time weight to 5.0 - do controls saturate the bounds?

2. Fixed time: Remove Δt from controls and compare total time.

3. Add waypoint: Require passing through [0.5, 0, 0.5] at the midpoint:

constraint = NonlinearKnotPointConstraint(
    x -> x - [0.5, 0, 0.5], :x, traj;
    times=[div(N,2)], equality=true
)
prob = DirectTrajOptProblem(traj, obj, integrator; constraints=[constraint])

4. Different goal: Try reaching [0, 1, 0] or [0.5, 0.5, 0.5]

5. Tighter bounds: Use bounds=(u = 0.5, Δt = (0.05, 0.3)) - how does time change?


This page was generated using Literate.jl.