stomp_motion_planner::PolicyImprovement Class Reference

Public Member Functions
	PolicyImprovement ()
	Constructor for the policy improvement class.
	~PolicyImprovement ()
	Destructor.
bool	initialize (const int num_rollouts, const int num_time_steps, const int num_reused_rollouts, const int num_extra_rollouts, boost::shared_ptr< stomp_motion_planner::Policy > policy, boost::shared_ptr< stomp_motion_planner::Task > task, bool use_cumulative_costs=true)
	Initializes the object which is required for all operations to succeed. More...
bool	setNumRollouts (const int num_rollouts, const int num_reused_rollouts, const int num_extra_rollouts)
	Resets the number of rollouts. More...
bool	getRollouts (std::vector< std::vector< Eigen::VectorXd > > &rollouts, const std::vector< double > &noise_stddev, bool get_reused, std::vector< std::vector< Eigen::VectorXd > > &reused_rollouts)
	Gets the next set of rollouts. More...
bool	setRolloutCosts (const Eigen::MatrixXd &costs, const double control_cost_weight, std::vector< double > &rollout_costs_total)
	Set the costs of each rollout per time-step Only the first "n" rows of the costs matrix is used, where n is the number of rollouts generated by getRollouts(), because some rollouts may be used from previous iterations. More...
bool	improvePolicy (std::vector< Eigen::MatrixXd > &parameter_updates)
	Performs the PI^2 update and provides parameter updates at every time step. More...
bool	addExtraRollouts (std::vector< std::vector< Eigen::VectorXd > > &rollouts, std::vector< Eigen::VectorXd > &rollout_costs)
	Adds extra rollouts to the set of rollouts to be reused.
void	testNoiseGenerators ()
	Tests the noise generators Function added by jim.
void	setRolloutOutOfBounds (int id, bool out_of_bounds)
	Set rollout as out of bounds Function added by jim.
bool	resetReusedRollouts ()
	Reset extra rollouts.

Public Attributes
std::vector< Eigen::MatrixXd >	projection_matrix_
	[num_dimensions] num_parameters x num_parameters

Member Function Documentation

bool stomp_motion_planner::PolicyImprovement::getRollouts	(	std::vector< std::vector< Eigen::VectorXd > > &	rollouts,
		const std::vector< double > &	noise_stddev,
		bool	get_reused,
		std::vector< std::vector< Eigen::VectorXd > > &	reused_rollouts
	)

Gets the next set of rollouts.

Only "new" rollouts that need to be executed are returned, not rollouts which might be reused from the previous set.

Parameters

rollouts_	[num_rollouts][num_dimensions] num_parameters
noise_variance	[num_dimensions] noise standard deviation per dimension

Returns

bool stomp_motion_planner::PolicyImprovement::improvePolicy

(

std::vector< Eigen::MatrixXd > &

parameter_updates

)

Performs the PI^2 update and provides parameter updates at every time step.

Parameters

parameter_updates

[num_dimensions] num_time_steps x num_parameters

Returns

bool stomp_motion_planner::PolicyImprovement::initialize	(	const int	num_rollouts,
		const int	num_time_steps,
		const int	num_reused_rollouts,
		const int	num_extra_rollouts,
		boost::shared_ptr< stomp_motion_planner::Policy >	policy,
		boost::shared_ptr< stomp_motion_planner::Task >	task,
		bool	use_cumulative_costs = true
	)

Initializes the object which is required for all operations to succeed.

Initialize policy improvment 1 - Allocates a multivariate gaussian sampler 2 - Computes the projection matrix.

Parameters

num_rollouts
num_time_steps
num_reused_rollouts
policy

Returns

true on success, false on failure

bool stomp_motion_planner::PolicyImprovement::setNumRollouts	(	const int	num_rollouts,
		const int	num_reused_rollouts,
		const int	num_extra_rollouts
	)

Resets the number of rollouts.

Allocates the rollout structures.

Parameters

num_rollouts

Returns

Parameters

num_rollouts	the total number of rollouts
num_reused_rollouts	the reused number of rollouts
num_extra_rollouts	the number of extra rollouts

bool stomp_motion_planner::PolicyImprovement::setRolloutCosts	(	const Eigen::MatrixXd &	costs,
		const double	control_cost_weight,
		std::vector< double > &	rollout_costs_total
	)

Set the costs of each rollout per time-step Only the first "n" rows of the costs matrix is used, where n is the number of rollouts generated by getRollouts(), because some rollouts may be used from previous iterations.

Outputs the total cost for each rollout (generated and reused) in rollout_costs_total

Parameters

costs

---

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

• src/planner/TrajectoryOptim/Stomp/policy_improvement.hpp
• src/planner/TrajectoryOptim/Stomp/policy_improvement.cpp