Search for grasp hypotheses. More...

#include <hand_search.h>

Public Member Functions
std::vector< GraspHypothesis >	findHands (const PointCloud::Ptr cloud, const Eigen::VectorXi &pts_cam_source, const std::vector< int > &indices, const PointCloud::Ptr cloud_plot, bool calculates_antipodal, bool uses_clustering)
	Find grasp hypotheses in a point cloud.
	HandSearch (double finger_width, double hand_outer_diameter, double hand_depth, double hand_height, double init_bite, int num_threads, int num_samples, bool plots_hands)
	Constructor.
Private Member Functions
std::vector< GraspHypothesis >	findHands (const PointCloud::Ptr cloud, const Eigen::VectorXi &pts_cam_source, const std::vector< Quadric > &quadric_list, const Eigen::VectorXi &hands_cam_source, const pcl::KdTreeFLANN< pcl::PointXYZ > &kdtree)
	Find grasp hypotheses in a point cloud given a list of quadratic surfaces.
std::vector< Quadric >	findQuadrics (const PointCloud::Ptr cloud, const Eigen::VectorXi &pts_cam_source, const pcl::KdTreeFLANN< pcl::PointXYZ > &kdtree, const std::vector< int > &indices)
	Find quadratic surfaces in a point cloud.
Private Attributes
Eigen::Matrix4d	cam_tf_left_
Eigen::Matrix4d	cam_tf_right_
	camera poses
Eigen::Matrix3Xd	cloud_normals_
	a 3xn matrix containing the normals for points in the point cloud
double	finger_width_
double	hand_depth_
	the finger length
double	hand_height_
	the hand extends plus/minus this value along the hand axis
double	hand_outer_diameter_
	the maximum robot hand aperture
double	init_bite_
	the minimum object height
double	nn_radius_hands_
	the radius for the neighborhood search for the hand search
double	nn_radius_taubin_
	the radius for the neighborhood search for the quadratic surface fit
int	num_samples_
	the number of samples used in the search
int	num_threads_
	the number of threads used in the search
Plot	plot_
	plot object for visualization of search results
bool	plots_camera_sources_
	whether the camera source for each point in the point cloud is plotted
bool	plots_hands_
	whether the grasp hypotheses are plotted
bool	plots_local_axes_
	whether the local axes estimated for each point neighborhood are plotted
bool	plots_samples_
	whether the samples drawn from the point cloud are plotted
bool	tolerant_antipodal_
	whether the antipodal testing uses "tolerant" thresholds
bool	uses_determinstic_normal_estimation_
	whether the normal estimation for the quadratic surface is deterministic (used for debugging)

Detailed Description

Search for grasp hypotheses.

HandSearch class

This class searches for grasp hypotheses in a point cloud by first fitting a quadratic surface to a small, local point neighborhood, and then finding geometrically feasible grasp hypotheses for a larger point neighborhood. It can also estimate whether the grasp is antipodal from the normals of the point neighborhood.

Definition at line 61 of file hand_search.h.

Constructor & Destructor Documentation

HandSearch::HandSearch	(	double	finger_width,
		double	hand_outer_diameter,
		double	hand_depth,
		double	hand_height,
		double	init_bite,
		int	num_threads,
		int	num_samples,
		bool	plots_hands
	)		`[inline]`

Constructor.

Parameters:

finger_width	the width of the robot hand fingers
hand_outer_diameter	the maximum robot hand aperture
hand_depth	the hand depth (length of fingers)
hand_height	the hand extends plus/minus this value along the hand axis
init_bite	the minimum object height
num_threads	the number of CPU threads to be used for the search
num_samples	the number of samples drawn from the point cloud
plots_hands	whether grasp hypotheses are plotted

Definition at line 76 of file hand_search.h.

Member Function Documentation

std::vector< GraspHypothesis > HandSearch::findHands	(	const PointCloud::Ptr	cloud,
		const Eigen::VectorXi &	pts_cam_source,
		const std::vector< int > &	indices,
		const PointCloud::Ptr	cloud_plot,
		bool	calculates_antipodal,
		bool	uses_clustering
	)

Find grasp hypotheses in a point cloud.

If the parameter indices is of size 0, then indices are randomly drawn from the point cloud according to a uniform distribution.

Parameters:

cloud	the point cloud that is searched for grasp hypotheses
pts_cam_source	the camera source for each point in the point cloud
indices	the list of point cloud indices that are used in the search
cloud_plot	the point cloud that is used for plotting
calculates_antipodal	whether grasp hypotheses are estimated to be antipodal
uses_clustering	whether clustering is used to divide the point cloud (useful for creating training data)

Returns:: a list of grasp hypotheses

Definition at line 4 of file hand_search.cpp.

std::vector< GraspHypothesis > HandSearch::findHands	(	const PointCloud::Ptr	cloud,
		const Eigen::VectorXi &	pts_cam_source,
		const std::vector< Quadric > &	quadric_list,
		const Eigen::VectorXi &	hands_cam_source,
		const pcl::KdTreeFLANN< pcl::PointXYZ > &	kdtree
	)		`[private]`

Find grasp hypotheses in a point cloud given a list of quadratic surfaces.

Parameters:

cloud	the point cloud that is searched for quadrics
pts_cam_source	the camera source for each point in the point cloud
quadric_list	the list of quadratic surfaces
hands_cam_source	the camera source for each sample
kdtree	the KD tree that is used for finding point neighborhoods

Returns:: a list of grasp hypotheses

Definition at line 116 of file hand_search.cpp.

std::vector< Quadric > HandSearch::findQuadrics	(	const PointCloud::Ptr	cloud,
		const Eigen::VectorXi &	pts_cam_source,
		const pcl::KdTreeFLANN< pcl::PointXYZ > &	kdtree,
		const std::vector< int > &	indices
	)		`[private]`

Find quadratic surfaces in a point cloud.

Parameters:

cloud	the point cloud that is searched for quadrics
pts_cam_source	the camera source for each point in the point cloud
kdtree	the KD tree that is used for finding point neighborhoods
indices	the list of point cloud indices that are used in the search