Simsflanagan Jl

SimsFlanagan.jl

Julia implementation of the Sims-Flanagan transcription method for low-thrust trajectory optimization. Repo: HAMMERHEAD-Space/SimsFlanagan.jl

Architecture

Spacecraft Types

AbstractSpacecraft
├── Spacecraft            # Constant thrust
├── SEPSpacecraft         # Solar electric propulsion (1/r^2 thrust scaling)
└── SolarSail             # Solar sail (radiation pressure)

Problem/Solution Types

SimsFlanaganProblem       # Problem definition (SciML pattern)
SimsFlanaganSolution      # Optimized trajectory result

Initial Guess Strategies

AbstractInitialGuess
├── RandomGuess           # Random throttle vectors
├── ZeroGuess             # Zero thrust (ballistic)
├── ConstantGuess         # Constant throttle direction
├── RadialGuess           # Radial thrust direction
└── LambertGuess          # Lambert arc-based initialization

Source Layout

src/
  SimsFlanagan.jl         # Module entry
  types.jl                # All type definitions
  problem.jl              # simsflanagan_problem() construction
  propagation.jl          # Kepler propagation, segment/leg propagation, mismatch
  solve.jl                # solve() via MadNLP optimizer
  utils.jl                # safe_norm (AD-safe), utilities

Key Patterns

SciML Interface

# Create problem
prob = simsflanagan_problem(r0, v0, rf, vf, tof, μ, spacecraft; kwargs...)

# Solve
sol = solve(prob; kwargs...)

# Remake for different parameters
prob2 = remake(prob; tof=new_tof)

Solution Accessors

position_mismatch(sol)     # Match-point position error
velocity_mismatch(sol)     # Match-point velocity error
mass_mismatch(sol)         # Mass continuity error

Optimization Stack

• MadNLP interior-point NLP solver with MUMPS linear solver
• ForwardDiff for gradient computation
• Optimization.jl unified interface via OptimizationMOI bridge

Trajectory Discretization

• Splits trajectory into forward/backward legs from match point
• Each segment has impulsive delta-V at midpoint (SVector{3} throttle)
• Sundman transformation for adaptive segment sizing based on orbital distance

AD-Safe Utilities

safe_norm(v)  # Handles zero vector without NaN gradients

Lambert-Based Initialization

Uses Lambert.jl to compute initial guess from Lambert arcs, providing better convergence than random initialization.

Adding a New Spacecraft Model

1. Define struct NewCraft <: AbstractSpacecraft with propulsion parameters
2. Implement thrust computation method (how thrust scales with position/state)
3. Add dispatch in propagation for the new thrust model
4. Add initial guess compatibility
5. Test: convergence on known transfer, AD gradient correctness

Dependencies

• AstroCoords 0.3, Lambert 0.1, SciMLBase 2, StaticArrays 1.9
• ForwardDiff 1.3 (AD for optimization gradients)
• MadNLP 0.8, MadNLPMumps 0.5 (interior-point NLP solver)
• Optimization 5, OptimizationMOI 1, OptimizationMadNLP 1