Skip to content

Robbins ​

  • 3 state
  • 1 control
  • Lagrange cost
  • fixed final time
  • 1 state box
  • 6 boundary
  • OptimalControl valid
  • OptimalControl, scalarized valid
  • JuMP valid
  • ModelingToolkit valid

The Robbins benchmark problem is a classical optimal control problem introduced by Robbins (1980), involving a third-order state constraint.

Mathematical formulation ​

The state is driven through its third derivative and must stay nonnegative:

with control and cost functional

The system is represented in first-order form as

with boundary conditions

and horizon  . In our simulations we use parameters  ,  ,  .

Qualitative behaviour ​

The exact solution exhibits infinitely many isolated contact points, where the state constraint   is active. Between these contact points, the unconstrained arcs decrease geometrically, accumulating at a point, followed by a trivial singular arc where   and  .

Numerically, capturing all contact points is challenging because the unconstrained arcs quickly shrink below the resolution of a given discretisation. Simulations reproduce the initial portion of the solution, showing the first few contact points. The control typically exhibits a bang–bang structure with possible singular arcs, while and evolve to satisfy the dynamics.

Characteristics ​

  • Third-order state constraint with inequality  ,

  • Linear dynamics in first-order representation,

  • Bang–bang control and singular arcs with infinitely many contact points,

  • Serves as a benchmark for direct transcription and NLP solvers handling state-constrained optimal control problems.

References ​

  • Robbins, H. M. (1980). Junction phenomena for optimal control with state-variable inequality constraints of third order. Journal of Optimization Theory and Applications, 31, 85–99. This is the original paper introducing the Robbins problem. It formulates the third-order state-constrained optimal control problem, describes the accumulation of contact points, and provides theoretical analysis of the junction phenomena.

  • Hermant, A. (2008). On the shooting algorithm for optimal control problems with state constraints (PhD thesis, École Polytechnique X). This thesis discusses numerical shooting methods for state-constrained optimal control problems, including the Robbins problem. It provides practical insights into solving problems with multiple contact points and complex singular arcs.

  • Jacobson, D. H., Lele, M. M., & Speyer, J. L. (1971). New necessary conditions of optimality for control problems with state-variable inequality constraints. Journal of Mathematical Analysis and Applications, 35, 255–284. This foundational work introduces necessary conditions for optimality in control problems with state-variable inequality constraints, forming the theoretical basis for analyzing and solving problems like Robbins.

  • BOCOP repository: Robbins problem Contains a practical implementation of the Robbins benchmark in the BOCOP optimal control framework. This resource allows testing and benchmarking direct transcription and NLP solvers on the Robbins problem, reproducing the first few contact points in numerical simulations.

Parameters ​

ParameterDefault valueUnitMeaning
t00–initial time
tf10–final time
α3–weight of x₁ in the cost
β0–weight of x₁² in the cost
γ0.5–weight of u² in the cost
x₁_l0–lower bound of x₁ (the third-order state constraint)
x₁_t01–initial value of x₁
x₂_t0-2–initial value of x₂
x₃_t00–initial value of x₃
x₁_tf0–final value of x₁
x₂_tf0–final value of x₂
x₃_tf0–final value of x₃

Every modeling takes other values with the keyword parameters, merged with these: robbins(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.

ModelingSchemeGridReference objectiveSolvesDistinct solutionsConfigurations agree
JuMP:trapeze50019.4666333381yes, within 2.1e-08
OptimalControl:trapeze50019.46663292181yes, within 8.3e-08
OptimalControl, scalarized:trapeze50019.4666332831yes, within 2.9e-12