19 Dec 10:07

6aeceee

v0.6.0 Latest

Latest

DirectTrajOpt v0.6.0

This release introduces a massive refactor of the internal API, standardizing how constraints, integrators, and objectives interact via a new Common Interface. It also brings compatibility with NamedTrajectories v0.7, shifting from index-based logic to symbolic component-based logic.

Breaking Changes

Common Interface Adoption: All Constraints, Integrators, and Objectives now implement the CommonInterface. Direct calls to internal evaluation functions must be updated to use the unified evaluate!, jacobian!, and hessian_of_lagrangian API.
Variable-Based Constraints: Constraints now use variable names (symbols) instead of raw integer indices for improved readability and robustness.
Problem Structure: DirectTrajOptProblem now stores integrators directly; the standalone dynamics.jl and its associated structures have been removed.
Trajectory Handling: Updated to NamedTrajectories v0.7. Functions that previously took raw vectors/indices now expect NamedTrajectory objects or symbolic references.

Added

CommonInterface Module: A unified dispatch system for evaluate!, jacobian_structure, jacobian!, hessian_structure, and hessian_of_lagrangian.
New Constraint Types:
Linear/Equality: AllEqualConstraint, TimeStepsAllEqualConstraint, EqualityConstraint, GlobalEqualityConstraint.
Boundary/Symmetry: BoundsConstraint, GlobalBoundsConstraint, SymmetricConstraint, SymmetricControlConstraint.
Aggregates: TotalConstraint, DurationConstraint (helper for time-summation).
Nonlinear Constraints: Added NonlinearKnotPointConstraint and NonlinearGlobalKnotPointConstraint for complex, non-linear trajectory requirements on knot-points.
Testing Utilities: Added test_constraint for validating nonlinear constraints against finite-difference approximations with detailed diff reporting.

Changes

Performance Optimization: Constraint and objective functions now compute component indices once at construction rather than at evaluation time.
Integrator Refactor: BilinearIntegrator, DerivativeIntegrator, and others now utilize NamedTrajectory for state dimension retrieval and symbolic component mapping.
Refined QuadraticRegularizer and MinimumTimeObjective for more efficient gradient and Hessian computations.
DirectTrajOptProblem now utilizes AbstractObjective for better extensibility.
IpoptEvaluator was updated to align with new dynamics creation and objective evaluation at runtime making it a bit more modular

Removed

src/dynamics.jl: Obsolete following the integration of dynamics directly into the problem and integrator structures.
Deprecated index-based constructors for constraints and objectives.
Certain objectives pre-allocated hessian matrices in various objective types that are being tested for internal use

Merged pull requests:

Bump codecov/codecov-action from 4 to 5 (#1) (@dependabot[bot])
tagbot + manual tag multidocs rebuild trigger (#36) (@jack-champagne)
Bump actions/upload-artifact from 4 to 5 (#37) (@dependabot[bot])
Bump actions/checkout from 4 to 5 (#38) (@dependabot[bot])
Bump peter-evans/repository-dispatch from 2 to 4 (#39) (@dependabot[bot])
refactor interface, new constraints, latest named trajectories (#41) (@jack-champagne)

Contributors

dependabot and jack-champagne

Assets 2

27 Oct 04:45

github-actions

v0.5.0

d5f3dbb

v0.5.0

DirectTrajOpt v0.5.0

Diff since v0.4.4

DirectTrajOpt.jl v0.5.0

Major documentation improvements and variable naming refactor for consistency.

Breaking Changes:

Trajectory time step variables renamed from T to N throughout (now consistent with standard notation for number of knot points)
Updated NamedTrajectories.jl dependency to v0.6

New Features:

Complete example added: time-optimal bilinear control with multiple inputs and free time steps
Visualization support via CairoMakie.jl integration in documentation
Enhanced quickstart guide with plotting examples

Documentation:

Comprehensive refactoring of all mathematical formulations to use N (knot points) convention
New detailed tutorials: linear system control, bilinear control, minimum time optimization
Enhanced concept guides for trajectories, integrators, objectives, and constraints

Improvements:

Julia 1.12 support
Refactored TimeDependentBilinearIntegrator to use OrdinaryDiffEqTsit5
Better alignment with mathematical optimization conventions

Assets 2

25 Oct 05:33

github-actions

v0.4.4

753c6bb

v0.4.4

DirectTrajOpt v0.4.4

Diff since v0.4.3

Added

support and CI for Julia 1.12
support for running nightly 'pre' for early testing

Merged pull requests:

support julia 1.12 (#33) (@jack-champagne)
add multidoc workflow trigger #98 (#34) (@jack-champagne)

Contributors

jack-champagne

Assets 2

24 Oct 18:44

github-actions

v0.4.3

910d9d0

v0.4.3

DirectTrajOpt v0.4.3

Diff since v0.4.2

Changes

New feature: DirectTrajOpt.Callbacks (reexported from DirectTrajOpt.IpoptSolverExt) implements i) an interface for generating and forwarding optimizer state to a callback, once per iteration of an Ipopt NLP, and ii) a handful of useful default factory routines for generating callbacks compatible with the foregoing interface.
- Note: Previous IpoptSolverExt implementations decoupled optimizer variables and NamedTrajectory state, with the consequence that early termination of solves resulted in loss of solver progress. Among the default factory routines implemented in Callbacks is a routine enforcing updates of NamedTrajectory state once per iteration, ameliorating this issue.

Merged pull requests: