Moonlander
- 6 state
- 2 control
- 1 variable
- Mayer cost
- free final time
- 2 control box
- 1 variable box
- 10 boundary
- OptimalControl valid
- JuMP valid
The Moonlander optimal control problem is a benchmark in constrained optimal control. It models the powered descent of a spacecraft aiming to land on a specified target on the Moon's surface in minimum time. The system comprises six state variables: the horizontal position
Mathematical formulation
The problem can be stated as
Here,
Qualitative behaviour
The optimal solution typically exhibits a bang–bang structure, where thrusts switch between minimum and maximum values, ensuring minimal landing time while satisfying position and velocity constraints. Orientation dynamics
Characteristics
Nonlinear coupled dynamics with orientation-dependent thrust.
Terminal position and velocity constraints for a soft landing.
Benchmark problem for testing direct transcription and NLP solvers.
References
Gazzola, F., & Marchini, E. M. (2021). The moon lander optimal control problem revisited. Mathematics in Engineering, 3(5), 1–14. doi:10.3934/mine.2021040 Provides a detailed analysis of the Moonlander problem, including minimum-time trajectories and safe landing curves.
GPOPS-II Examples: Moon Lander Problem. GPOPS-II Moon Lander Example Illustrates practical implementation of the Moonlander OCP in a direct transcription framework, useful for benchmarking.
Vanroye, L., Sathya, A., De Schutter, J., & Decré, W. (2023). FATROP: A Fast Constrained Optimal Control Problem Solver. arXiv preprint arXiv:2303.16746. Discusses solver strategies applied to constrained trajectory optimization problems, including applications relevant to the Moonlander.
Parameters
| Parameter | Default value | Unit | Meaning |
|---|---|---|---|
t0 | 0 | s | initial time |
m | 1 | kg | mass of the lander |
g | 9.81 | m/s² | gravitational acceleration |
I | 0.1 | kg·m² | moment of inertia of the lander |
D | 1 | m | distance between the two thrusters |
max_thrust | 19.62 | N | upper bound of each thrust F₁, F₂ |
tf_l | 0.1 | s | lower bound of the final time |
tf_u | 1.0 | s | upper bound of the final time |
F₁_l | 0 | N | lower bound of the thrust F₁ |
F₂_l | 0 | N | lower bound of the thrust F₂ |
p₁_t0 | 0 | m | initial horizontal position |
p₂_t0 | 0 | m | initial vertical position |
dp₁_t0 | 0 | m/s | initial horizontal velocity |
dp₂_t0 | 0 | m/s | initial vertical velocity |
θ_t0 | 0 | rad | initial orientation |
dθ_t0 | 0 | rad/s | initial angular velocity |
p₁_tf | 5 | m | final horizontal position, the target |
p₂_tf | 5 | m | final vertical position, the target |
dp₁_tf | 0 | m/s | final horizontal velocity |
dp₂_tf | 0 | m/s | final vertical velocity |
Every modeling takes other values with the keyword parameters, merged with these: moonlander(backend; parameters=(t0=…,)).
Reference objective
The best objective among the solves that converged, in each modeling, on the default grid of the modeling: the modelings are independent, each has its own reference. See how the references are computed.
| Modeling | Scheme | Grid | Reference objective | Solves | Distinct solutions | Configurations agree |
|---|---|---|---|---|---|---|
| JuMP | :trapeze | 500 | 0.9618109334 | 5 | 1 | yes, within 9.2e-07 |
| OptimalControl | :trapeze | 500 | 0.9618109334 | 17 | 2 | no |
In OptimalControl with :trapeze, the solves reached 2 distinct solutions: besides the best one, 0.9618437656 (gap 3.4e-05, 1 solves). See the results of the modeling for which configurations reached which.