ENH: Control and Guidance to land a rocket on the ground (GFOLD) #209

Activity

1. giovaniceotto

   added this to RocketPy action logon Aug 20, 2022
2. giovaniceotto

   moved this to Todo in RocketPy action logon Aug 20, 2022
3. Gui-FernandesBR

   added

   Help wantedExtra attention is neededExtra attention is needed

   on Sep 5, 2022
4. Gui-FernandesBR

   added this to the RocketPy at EuroC 2022 milestone on Sep 20, 2022
5. Gui-FernandesBR

   added

   ControllersControlling rocket flight methodsControlling rocket flight methods

   on Sep 21, 2022
6. Gui-FernandesBR

   removed this from the RocketPy at EuroC 2022 milestone on Oct 4, 2022
7. Gui-FernandesBR

   added this to the Release v2.0.0 milestone on Nov 15, 2022
8. MateusStano

   added this to LibDev Roadmapon Nov 4, 2023
9. github-project-automation

   moved this to v1.X in LibDev Roadmapon Nov 4, 2023
10. Gui-FernandesBR

    changed the title [-]Control and Guidance[/-] [+]ENH: Control and Guidance to land a rocket on the ground (GFOLD)[/+] on Nov 2, 2024
11. Gui-FernandesBR

    added

    EnhancementNew feature or request, including adjustments in current codesNew feature or request, including adjustments in current codes

    on Nov 2, 2024
12. Gui-FernandesBR

    added the

    Feature

    issue type on Mar 4, 2025
13. hahahuy commented on Apr 8, 2026

    

    hahahuy

    on Apr 8, 2026

    More actions

    Relationship to issue #952 (powered descent physics)

    I've opened issue #952 which proposes the simulation physics layer that GFOLD (this issue) would need as a foundation:

    #952 — Proposal: powered descent / retropropulsion module
    #952

    The split in scope:

    	#952 (physics)	#209 (guidance — this issue)
    What it adds	New u_dot_powered_descent ODE derivative, descent_motor phase in FlightPhases, motor mass accounting during descent	GFOLD fuel-optimal guidance law (CasADi or custom solver)
    Dependency	None — standalone physics	Requires #952 to be testable end-to-end
    Integration	New Flight parameters: descent_motor, descent_trigger, guidance_schedule	Produces the guidance_schedule (thrust vs. time) that #952 replays

    The key design decision discussed in #952: guidance is pre-solved offline and replayed as a thrust schedule during integration — no optimizer inside the ODE loop. This keeps Monte Carlo runs fast (~0.1 s/trajectory) while still supporting optimal landing.

    Would love to coordinate if there's active work on #209 to make sure the physics API (#952) exposes what the guidance layer needs.
14. ting-hong-shieh commented on Aug 14, 2026

    

    ting-hong-shieh

    on Aug 14, 2026

    More actions

    The dynamics prerequisite is now being discussed in #952. I suggest keeping GFOLD as a guidance layer that produces a schedule, with RocketPy's powered-descent implementation responsible for executing that schedule.

    The first interface can be solver-neutral:

    • input: initial state, target state, vehicle limits, thrust bounds, mass model, and time bounds;
    • output: timestamped thrust magnitude/direction and ignition/shutdown schedule; and
    • status: optimal, infeasible, iteration/solver failure, with diagnostics that do not depend on one optimizer.

    CasADi or another optimizer can remain an optional backend behind that interface. The core package should still be able to import and run ordinary Flight simulations without the optimization dependency installed.

    Acceptance should include one fixed feasible benchmark with a known objective range, one infeasible constraint set, schedule serialization, and replay of the schedule through the #952 dynamics to verify the terminal state within stated tolerances. Guidance tests should not pass merely because the optimizer returned success; the replayed physics result is the relevant end-to-end check.

    I recommend waiting until #952 has a stable event and thrust-command API before choosing public GFOLD classes. Once that prerequisite exists, I can help define the benchmark fixture and schedule protocol as a separate first milestone.

    👍React with 👍1Reacted by Gui