The feasible DDP (FDDP) algorithm, from Mastalli et al. (2020). More...

#include <aligator/solvers/fddp/solver-fddp.hpp>

Collaboration diagram for aligator::SolverFDDPTpl< Scalar >:

Public Types
using	Problem = TrajOptProblemTpl<Scalar>
using	StageModel = StageModelTpl<Scalar>
using	StageData = StageDataTpl<Scalar>
using	ProblemData = TrajOptDataTpl<Scalar>
using	Results = ResultsFDDPTpl<Scalar>
using	Workspace = WorkspaceFDDPTpl<Scalar>
using	Manifold = ManifoldAbstractTpl<Scalar>
using	VParams = ValueFunctionTpl<Scalar>
using	QParams = QFunctionTpl<Scalar>
using	CostData = CostDataAbstractTpl<Scalar>
using	DynamicsModel = ExplicitDynamicsModelTpl<Scalar>
using	ExplicitDynamicsData = ExplicitDynamicsDataTpl<Scalar>
using	CallbackPtr = shared_ptr<CallbackBaseTpl<Scalar>>
using	CallbackMap = boost::unordered_map<std::string, CallbackPtr>

Public Member Functions
	ALIGATOR_DYNAMIC_TYPEDEFS (Scalar)
void	setNumThreads (const std::size_t num_threads)
std::size_t	getNumThreads () const
	SolverFDDPTpl (const Scalar tol=1e-6, VerboseLevel verbose=VerboseLevel::QUIET, const Scalar reg_init=1e-9, const std::size_t max_iters=200)
void	setup (const Problem &problem)
	Allocate workspace and results structs.
void	updateExpectedImprovement (Workspace &workspace, Results &results) const
	Pre-compute parts of the directional derivatives – this is done before linesearch.
void	expectedImprovement (Workspace &workspace, Scalar &d1, Scalar &d2) const
	Finish computing the directional derivatives – this is done within linesearch.

Static Public Member Functions
static Scalar	forwardPass (const Problem &problem, const Results &results, Workspace &workspace, const Scalar alpha)
	Perform a nonlinear rollout, keeping an infeasibility gap.

Public Attributes
Scalar	target_tol_
Scalar	th_grad_ = 1e-12
Scalar	th_step_dec_ = 0.5
Scalar	th_step_inc_ = 0.01
Linesearch< Scalar >::Options	ls_params
VerboseLevel	verbose_
std::size_t	max_iters
	Maximum number of iterations for the solver.
bool	force_initial_condition_
Logger	logger {}
Results	results_
Workspace	workspace_
Regularization parameters
Scalar	reg_init
Scalar	preg_ = reg_init
Scalar	reg_min_ = 1e-9
Scalar	reg_max_ = 1e9
Scalar	reg_dec_factor_ = 0.1
	Regularization decrease factor.
Scalar	reg_inc_factor_ = 10.
	Regularization increase factor.

Protected Attributes
std::size_t	num_threads_
	Number of threads to use when evaluating the problem or its derivatives.
CallbackMap	callbacks_
	Callbacks.

Detailed Description

template<typename Scalar>
struct aligator::SolverFDDPTpl< Scalar >

The feasible DDP (FDDP) algorithm, from Mastalli et al. (2020).

The implementation very similar to Crocoddyl's SolverFDDP.

Definition at line 27 of file solver-fddp.hpp.

Member Typedef Documentation

◆ Problem

template<typename Scalar>

using aligator::SolverFDDPTpl< Scalar >::Problem = TrajOptProblemTpl<Scalar>

Definition at line 29 of file solver-fddp.hpp.

◆ StageModel

template<typename Scalar>

using aligator::SolverFDDPTpl< Scalar >::StageModel = StageModelTpl<Scalar>

Definition at line 30 of file solver-fddp.hpp.

◆ StageData

template<typename Scalar>

using aligator::SolverFDDPTpl< Scalar >::StageData = StageDataTpl<Scalar>

Definition at line 31 of file solver-fddp.hpp.

◆ ProblemData

template<typename Scalar>

using aligator::SolverFDDPTpl< Scalar >::ProblemData = TrajOptDataTpl<Scalar>

Definition at line 32 of file solver-fddp.hpp.

◆ Results

template<typename Scalar>

using aligator::SolverFDDPTpl< Scalar >::Results = ResultsFDDPTpl<Scalar>

Definition at line 33 of file solver-fddp.hpp.

◆ Workspace

template<typename Scalar>

using aligator::SolverFDDPTpl< Scalar >::Workspace = WorkspaceFDDPTpl<Scalar>

Definition at line 34 of file solver-fddp.hpp.

◆ Manifold

template<typename Scalar>

using aligator::SolverFDDPTpl< Scalar >::Manifold = ManifoldAbstractTpl<Scalar>

Definition at line 35 of file solver-fddp.hpp.

◆ VParams

template<typename Scalar>

using aligator::SolverFDDPTpl< Scalar >::VParams = ValueFunctionTpl<Scalar>

Definition at line 36 of file solver-fddp.hpp.

◆ QParams

template<typename Scalar>

using aligator::SolverFDDPTpl< Scalar >::QParams = QFunctionTpl<Scalar>

Definition at line 37 of file solver-fddp.hpp.

◆ CostData

template<typename Scalar>

using aligator::SolverFDDPTpl< Scalar >::CostData = CostDataAbstractTpl<Scalar>

Definition at line 38 of file solver-fddp.hpp.

◆ DynamicsModel

template<typename Scalar>

using aligator::SolverFDDPTpl< Scalar >::DynamicsModel = ExplicitDynamicsModelTpl<Scalar>

Definition at line 39 of file solver-fddp.hpp.

◆ ExplicitDynamicsData

template<typename Scalar>

using aligator::SolverFDDPTpl< Scalar >::ExplicitDynamicsData = ExplicitDynamicsDataTpl<Scalar>

Definition at line 40 of file solver-fddp.hpp.

◆ CallbackPtr

template<typename Scalar>

using aligator::SolverFDDPTpl< Scalar >::CallbackPtr = shared_ptr<CallbackBaseTpl<Scalar>>

Definition at line 41 of file solver-fddp.hpp.

◆ CallbackMap

template<typename Scalar>

using aligator::SolverFDDPTpl< Scalar >::CallbackMap = boost::unordered_map<std::string, CallbackPtr>

Definition at line 42 of file solver-fddp.hpp.

Constructor & Destructor Documentation

◆ SolverFDDPTpl()

template<typename Scalar>

aligator::SolverFDDPTpl< Scalar >::SolverFDDPTpl	(	const Scalar	tol = 1e-6,
		VerboseLevel	verbose = VerboseLevel::QUIET,
		const Scalar	reg_init = 1e-9,
		const std::size_t	max_iters = 200 )

Member Function Documentation

◆ ALIGATOR_DYNAMIC_TYPEDEFS()

template<typename Scalar>

aligator::SolverFDDPTpl< Scalar >::ALIGATOR_DYNAMIC_TYPEDEFS ( Scalar )

◆ setNumThreads()

template<typename Scalar>

void aligator::SolverFDDPTpl< Scalar >::setNumThreads ( const std::size_t num_threads )

inline

Definition at line 74 of file solver-fddp.hpp.

◆ getNumThreads()

template<typename Scalar>

std::size_t aligator::SolverFDDPTpl< Scalar >::getNumThreads ( ) const

inline

Definition at line 78 of file solver-fddp.hpp.

◆ setup()

template<typename Scalar>

void aligator::SolverFDDPTpl< Scalar >::setup ( const Problem & problem )

Allocate workspace and results structs.

◆ forwardPass()

template<typename Scalar>

Scalar aligator::SolverFDDPTpl< Scalar >::forwardPass	(	const Problem &	problem,
		const Results &	results,
		Workspace &	workspace,
		const Scalar	alpha )

static

Perform a nonlinear rollout, keeping an infeasibility gap.

Perform a nonlinear rollout using the computed sensitivity gains from the backward pass, while keeping the dynamical feasibility gaps open proportionally to the step-size alpha.

Parameters

[in]	problem
[in]	results
[out]	workspace
[in]	alpha	step-size.

◆ updateExpectedImprovement()

template<typename Scalar>

void aligator::SolverFDDPTpl< Scalar >::updateExpectedImprovement	(	Workspace &	workspace,
		Results &	results ) const

Pre-compute parts of the directional derivatives – this is done before linesearch.

Inspired from Crocoddyl's own function, crocoddyl::SolverFDDP::updateExpectedImprovement

◆ expectedImprovement()

template<typename Scalar>

void aligator::SolverFDDPTpl< Scalar >::expectedImprovement	(	Workspace &	workspace,
		Scalar &	d1,
		Scalar &	d2 ) const

Finish computing the directional derivatives – this is done within linesearch.

Inspired from Crocoddyl's own function, crocoddyl::SolverFDDP::expectedImprovement

Member Data Documentation

◆ target_tol_

template<typename Scalar>

Scalar aligator::SolverFDDPTpl< Scalar >::target_tol_

Definition at line 44 of file solver-fddp.hpp.

◆ reg_init

template<typename Scalar>

Scalar aligator::SolverFDDPTpl< Scalar >::reg_init

Definition at line 48 of file solver-fddp.hpp.

◆ preg_

template<typename Scalar>

Scalar aligator::SolverFDDPTpl< Scalar >::preg_ = reg_init

Definition at line 49 of file solver-fddp.hpp.

◆ reg_min_

template<typename Scalar>

Scalar aligator::SolverFDDPTpl< Scalar >::reg_min_ = 1e-9

Definition at line 50 of file solver-fddp.hpp.

◆ reg_max_

template<typename Scalar>

Scalar aligator::SolverFDDPTpl< Scalar >::reg_max_ = 1e9

Definition at line 51 of file solver-fddp.hpp.

◆ reg_dec_factor_

template<typename Scalar>

Scalar aligator::SolverFDDPTpl< Scalar >::reg_dec_factor_ = 0.1

Regularization decrease factor.

Definition at line 53 of file solver-fddp.hpp.

◆ reg_inc_factor_

template<typename Scalar>

Scalar aligator::SolverFDDPTpl< Scalar >::reg_inc_factor_ = 10.

Regularization increase factor.

Definition at line 55 of file solver-fddp.hpp.

◆ th_grad_

template<typename Scalar>

Scalar aligator::SolverFDDPTpl< Scalar >::th_grad_ = 1e-12

Definition at line 58 of file solver-fddp.hpp.

◆ th_step_dec_

template<typename Scalar>

Scalar aligator::SolverFDDPTpl< Scalar >::th_step_dec_ = 0.5

Definition at line 59 of file solver-fddp.hpp.

◆ th_step_inc_

template<typename Scalar>

Scalar aligator::SolverFDDPTpl< Scalar >::th_step_inc_ = 0.01

Definition at line 60 of file solver-fddp.hpp.

◆ ls_params

template<typename Scalar>

Linesearch<Scalar>::Options aligator::SolverFDDPTpl< Scalar >::ls_params

Definition at line 62 of file solver-fddp.hpp.

◆ verbose_

template<typename Scalar>

VerboseLevel aligator::SolverFDDPTpl< Scalar >::verbose_

Definition at line 64 of file solver-fddp.hpp.

◆ max_iters

template<typename Scalar>

std::size_t aligator::SolverFDDPTpl< Scalar >::max_iters

Maximum number of iterations for the solver.

Definition at line 66 of file solver-fddp.hpp.

◆ force_initial_condition_

template<typename Scalar>

bool aligator::SolverFDDPTpl< Scalar >::force_initial_condition_

Crocoddyl's FDDP implementation forces the initial state in linesearch to satisfy the initial condition. This flag switches that behaviour on or off.

Definition at line 70 of file solver-fddp.hpp.

◆ logger

template<typename Scalar>

Logger aligator::SolverFDDPTpl< Scalar >::logger {}

Definition at line 72 of file solver-fddp.hpp.

◆ num_threads_

template<typename Scalar>

std::size_t aligator::SolverFDDPTpl< Scalar >::num_threads_

protected

Number of threads to use when evaluating the problem or its derivatives.

Definition at line 82 of file solver-fddp.hpp.

◆ callbacks_

template<typename Scalar>

CallbackMap aligator::SolverFDDPTpl< Scalar >::callbacks_

protected

Callbacks.

Definition at line 84 of file solver-fddp.hpp.

◆ results_

template<typename Scalar>

Results aligator::SolverFDDPTpl< Scalar >::results_

Definition at line 87 of file solver-fddp.hpp.

◆ workspace_

template<typename Scalar>

Workspace aligator::SolverFDDPTpl< Scalar >::workspace_

Definition at line 88 of file solver-fddp.hpp.

The documentation for this struct was generated from the following files:

include/aligator/fwd.hpp
include/aligator/solvers/fddp/solver-fddp.hpp

Public Types

Public Member Functions

Static Public Member Functions

Public Attributes

Protected Attributes

Detailed Description

Member Typedef Documentation

◆ Problem

◆ StageModel

◆ StageData

◆ ProblemData

◆ Results

◆ Workspace

◆ Manifold

◆ VParams

◆ QParams

◆ CostData

◆ DynamicsModel

◆ ExplicitDynamicsData

◆ CallbackPtr

◆ CallbackMap

Constructor & Destructor Documentation

◆ SolverFDDPTpl()

Member Function Documentation

◆ ALIGATOR_DYNAMIC_TYPEDEFS()

◆ setNumThreads()

◆ getNumThreads()

◆ setup()

◆ forwardPass()

◆ updateExpectedImprovement()

◆ expectedImprovement()

Member Data Documentation

◆ target_tol_

◆ reg_init

◆ preg_

◆ reg_min_

◆ reg_max_

◆ reg_dec_factor_

◆ reg_inc_factor_

◆ th_grad_

◆ th_step_dec_

◆ th_step_inc_

◆ ls_params

◆ verbose_

◆ max_iters

◆ force_initial_condition_

◆ logger

◆ num_threads_

◆ callbacks_

◆ results_

◆ workspace_