simple_casadi_mpc::JITMPC Class Reference

MPC variant that JIT-compiles the solver during construction. More...

#include <simple_casadi_mpc.hpp>

Inheritance diagram for simple_casadi_mpc::JITMPC:

Public Member Functions
template<class T >
	JITMPC (const std::string &prob_name, std::shared_ptr< T > prob, std::string solver_name="ipopt", casadi::Dict config=MPC::default_ipopt_config(), casadi::Dict jit_options=JITMPC::default_jit_options(), const bool verbose=false)
	Build the NLP and JIT-compile its solver in one step.
Eigen::VectorXd	solve (Eigen::VectorXd x0, casadi::DMDict new_param_list=casadi::DMDict()) override
	Solve the NLP at the current state and return the first optimal control.
Public Member Functions inherited from simple_casadi_mpc::MPC
template<class T >
	MPC (std::shared_ptr< T > prob, std::string solver_name="ipopt", casadi::Dict config=default_ipopt_config())
	Build the NLP from prob and create the underlying nlpsol.
casadi::MXDict	casadi_prob () const
	Symbolic NLP description {x, f, g, p} constructed from the Problem.
const std::string &	solver_name () const
	Backend name passed to CasADi nlpsol.
casadi::Dict	solver_config () const
	Effective config forwarded to nlpsol (after consuming simple-casadi-mpc keys).
std::vector< casadi_int >	equality_flags () const
	Per-constraint flags marking equality (true) vs inequality (false).

Static Public Member Functions
static casadi::Dict	default_jit_options ()
	Default JIT compile options (compiler / flags / verbose) passed to CasADi.
Static Public Member Functions inherited from simple_casadi_mpc::MPC
static casadi::Dict	default_ipopt_config ()
	Reasonable defaults for the IPOPT backend (silent, warm-start enabled).
static casadi::Dict	default_qpoases_config ()
	Reasonable defaults for SQP method with the qpOASES inner QP solver.
static casadi::Dict	default_fatrop_config ()
	Reasonable defaults for the FATROP backend (auto structure detection).
static bool	equality_required (const std::string &solver_name, const casadi::Dict &config)
	Whether the chosen backend needs an equality flag vector in the config.

Additional Inherited Members
Protected Member Functions inherited from simple_casadi_mpc::MPC
void	build_with_map (size_t nx, size_t nu, casadi_int N, bool mapsum_stage_cost=true, bool expand_inner_functions=true)
Protected Attributes inherited from simple_casadi_mpc::MPC
std::shared_ptr< Problem >	prob_
std::string	solver_name_
casadi::Dict	config_
casadi::MXDict	casadi_prob_
casadi::Function	solver_
std::vector< casadi::MX >	Xs = {}
std::vector< casadi::MX >	Us = {}
casadi::DM	lbw_
casadi::DM	ubw_
casadi::DM	lbg_
casadi::DM	ubg_
std::vector< casadi::DM >	param_vec_ = {}
std::vector< bool >	equality_ = {}
casadi::DM	w0_
casadi::DM	lam_x0_
casadi::DM	lam_g0_

Detailed Description

MPC variant that JIT-compiles the solver during construction.

Pays a one-time compile cost (cache it with ccache via default_jit_options()) in exchange for substantially faster per-iteration solve time. Behaves like MPC otherwise.

Constructor & Destructor Documentation

JITMPC()

template<class T >

simple_casadi_mpc::JITMPC::JITMPC ( const std::string & prob_name, std::shared_ptr< T > prob, std::string solver_name = "ipopt", casadi::Dict config = MPC::default_ipopt_config(), casadi::Dict jit_options = JITMPC::default_jit_options(), const bool verbose = false )

inline

Build the NLP and JIT-compile its solver in one step.

Parameters

prob_name	unique identifier embedded in the JIT artifact name.
prob	the Problem to solve.
solver_name	CasADi nlpsol backend name.
config	nlpsol options (same semantics as MPC::MPC).
jit_options	inner CasADi jit_options dict; see default_jit_options.
verbose	if true, print progress to stdout while compiling.

Member Function Documentation

default_jit_options()

static casadi::Dict simple_casadi_mpc::JITMPC::default_jit_options ( )

inlinestatic

Default JIT compile options (compiler / flags / verbose) passed to CasADi.

The defaults are ccache gcc, -O3 -march=native, verbose=false. Override individual entries before passing the dict to the constructor:

auto opts = JITMPC::default_jit_options();
opts["compiler"] = "clang";
opts["flags"] = "-O2";
JITMPC mpc("my_prob", prob, "ipopt", config, opts);

solve()

Eigen::VectorXd simple_casadi_mpc::JITMPC::solve ( Eigen::VectorXd x0, casadi::DMDict new_param_list = casadi::DMDict() )

inlineoverridevirtual

Solve the NLP at the current state and return the first optimal control.

Warm-starts from the previous solve (x, lam_x, lam_g are cached internally), so calling this repeatedly during closed-loop simulation benefits from incremental convergence.

Parameters

x0	current measured state, shape (nx,).
new_param_list	updates to parameters declared via Problem::parameter. Keys are parameter names; values are casadi::DM of matching shape.

Returns

optimal control to apply now, u_0, shape (nu,).

Reimplemented from simple_casadi_mpc::MPC.

---

The documentation for this class was generated from the following file:

• /home/runner/work/simple_casadi_mpc/simple_casadi_mpc/include/simple_casadi_mpc/simple_casadi_mpc.hpp