Page cover image

Smart Parallel Picking Skill

Detection of different objects.

Overview

A typical mechanical type of gripping is the parallel grip, in which the mostly linear stroke movement of the gripper jaws enables objects to be lifted or gripped.

The possible gripping axes of the gripper jaws are represented in the GUI by the lines on the objects. There are three different color codes, blue describes the current gripping candidate. Yellow stands for possible other gripping points, while red indicates that these gripping axes are rejected. Depending on the settings, more or less probable grab possibilities are shown here. Limiting the possible grab point options allows an individually adjustable balance between quality and speed.

Similar to the "Smart Vacuum Picking Skill", this skill can be optimally used for separation tasks. It can be used in many ways after it recognizes the best picking candidate on unknown parts in the source box. Classically, it is used in logistics areas, where the variety of objects is often not very clear. It solves the problem of kitting in the automotive industry, as well as the separation in warehouse processes.

Cover

Boxes

Cover

Cylinders

Cover

M10 nuts

Cover

Random Items

Skill Result Information

Position

Position of the grasp point relative to the robot’s coordinate system (in m)

Orientation

Rotation of the grasp point relative to the robot’s coordinate system (in m)

Width

Determined grip width (in m)

Specifications

Conditions

Camera mount:

  • Dynamic

  • Static

Camera distance:

35 – 45 cm

Parts dimensions:

2 – 6 cm

Parts material:

matt - low reflection

Specs

Avg. recognition time:

< 0.5 seconds

Supported grippers:

  • Two-finger-gripper

  • Robotiq 2F-85

  • Zimmer GEH6040IL

  • OnRobot

Features

  • Collision detection

  • Generalized to a variety of object types

  • Picking from bulk

Parameter Example

To ensure accurate identification of various types of objects, the skill parameters can be easily adjusted to fit your specific needs. Here are some recommendations to help you find the perfect parameters for your application. For your convenience, we've only included descriptions of parameters that differ from the default settings.

Smart Parallel Picking Skill for boxes
  • Used objects: small cardboard boxes (70x50x50mm), matt surface

  • Camera distance: 550mm

  • Camera mount: 30° angled

  • Skill parameters:

    • Min score: 0.5

    • other parameters: default values

Technical Parameter Description

Parameter

Name

Edge sensitivity

Parameter

edge_sensitivity

Type

float

Description

Determines the sensitivity for the edge-detection algorithm. A larger sensitivity can help to detect small steps in depth values. However, a higher sensitivity also leads to more false edge detections and a longer runtime.

Name

Min score

Parameter

min_score

Type

float

Description

Minimum quality score for grasp candidates. The quality score is estimated by a neural network.

Name

Max candidates

Parameter

max_candidates

Type

int

Description

The maximum number of grasp candidates that are evaluated by the neural network. While larger numbers increase the runtime, they can also improve the robustness of the skill.

Name

Friction coefficient

Parameter

friction_coefficient

Type

float

Description

The friction coefficient between the gripper fingers and items. The default value of 0.2 fits for most materials.

Name

Min grasp distance

Parameter

min_grasp_distance

Type

float

Description

The minimum distance between two grasp candidates. This parameter is used to limit the number of very similar candidates and, thus, increasing the robustness and reducing the runtime of the skill.

Name

Grasp offset

Parameter

grasp_offset

Type

float

Description

An offset for the grasp width. This parameter is used to increase the predicted grasp with to avoid collisions between the gripper and the item to be picked.

Detections

Type

pose (Transformation)

Description

The pick pose for the parallel gripper

Type

width (float)

Description

Grasp width in meters

Type

quality (float)

Description

Grasp quality as estimated by the neural network.

Last updated