Generate Grasp Pose from Detection Results

Function: Converts detection results to pick object poses. Converts detection results from deep learning methods, combined with point cloud information, into pick object poses. [cite: 6]

Input Parameters:

Name	Type	Valid Range	Default Value	Meaning
Detection results	DetectInstance	None	{}	Returns bounding box, category, score, and polygon. [cite: 6]
Image	Image	None	None	Image, input can be grayscale or color. [cite: 6]
Camera coordinate system point cloud	XYZPoints	None	None	Point cloud. Since it’s converted using intrinsic parameters, it needs to be a point cloud in the camera coordinate system and must be an ordered point cloud to align with the image. [cite: 6]
Camera internal reference	Matrix	None	None	Camera intrinsic matrix. [cite: 6]
Camera distortion	List	None	None	Camera distortion coefficient vector. [cite: 6]
Hand-eye calibration matrix	Matrix	None	[[1. 0. 0. 0.] [0. 1. 0. 0.] [0. 0. 1. 0.] [0. 0. 0. 1.]]	Hand-eye calibration matrix, defaults to identity matrix. [cite: 6]

Name

Type

Valid Range

Default Value

Meaning

Detection results

DetectInstance

None

{}

Returns bounding box, category, score, and polygon. [cite: 6]

Image

None

Image, input can be grayscale or color. [cite: 6]

Camera coordinate system point cloud

XYZPoints

None

Point cloud. Since it’s converted using intrinsic parameters, it needs to be a point cloud in the camera coordinate system and must be an ordered point cloud to align with the image. [cite: 6]

Camera internal reference

Matrix

None

Camera intrinsic matrix. [cite: 6]

Camera distortion

List

None

Camera distortion coefficient vector. [cite: 6]

Hand-eye calibration matrix

Matrix

None

[[1. 0. 0. 0.] [0. 1. 0. 0.] [0. 0. 1. 0.] [0. 0. 0. 1.]]

Hand-eye calibration matrix, defaults to identity matrix. [cite: 6]

Output Parameters:

Name	Type	Valid Range	Default Value	Meaning
Grab Positional Information	PosesList	None	[]	Grab Positional Information. [cite: 6]

Name

Type

Valid Range

Default Value

Meaning

Grab Positional Information

PosesList

None

[]

Grab Positional Information. [cite: 6]

Parameter Settings:

Name	Type	Valid Range	Default Value	Meaning
Circle radius scale	Float	[0, 1]	0.3	Circle radius ratio, relative to half the short side of the bounding box, used to select object surface point cloud for depth calculation. [cite: 6]
Calculation method of the z value of the capture point	String	['Mean', 'Median']	Mean	The Z-value of the pick point is generally calculated using the mean. However, in scenarios with many outliers in the point cloud, using the median might avoid inaccuracies caused by outliers. [cite: 6]
Use detection result angle	Bool	None	False	If enabled, use the angle from the detection result to calculate the resulting pose rotation. If disabled, use the angular relationship between the robot coordinate system point cloud corresponding to the long side of the detection result’s minimum bounding box and the coordinate system axis for Rz rotation. [cite: 6]

Name

Type

Valid Range

Default Value

Meaning

Circle radius scale

Float

[0, 1]

0.3

Circle radius ratio, relative to half the short side of the bounding box, used to select object surface point cloud for depth calculation. [cite: 6]

Calculation method of the z value of the capture point

String

['Mean', 'Median']

Mean

The Z-value of the pick point is generally calculated using the mean. However, in scenarios with many outliers in the point cloud, using the median might avoid inaccuracies caused by outliers. [cite: 6]

Use detection result angle

Bool

None

False

If enabled, use the angle from the detection result to calculate the resulting pose rotation. If disabled, use the angular relationship between the robot coordinate system point cloud corresponding to the long side of the detection result’s minimum bounding box and the coordinate system axis for Rz rotation. [cite: 6]