Glossary¶
The terms used across the Pick[+] interface and this documentation, grouped by the part of the workflow they belong to.
Data Generation¶
| Term | Definition | Why it matters |
|---|---|---|
| Reference | An object or part variant registered in Data Generation, built from one or more scans and 3D models | It is the class the AI model learns to recognize, so Similarity-Based Picking can tell one part from another |
| Scan | The data captured when scanning a physical object with the Pick[+] camera. Depending on the camera and the trigger type, it holds 2D data, 3D data or both | It is the dataset behind everything else: annotations, 3D models and reference matching |
| Annotations | The data generated from a scan: Label, Mask, Pose and Embeddings | A reference can only be enabled once one of its scans has its mask and embeddings generated |
| Mask | Instance segmentation outline that isolates the scanned object from the background | Needed for most workflows |
| Embeddings | Represents the object's visual features, such as its shape, color, texture and surface pattern, extracted from the scan | What lets Pick[+] match a detected object against your references |
| 3D Model | The object's 3D geometry, either generated from one of its scans or imported as a CAD file | Required to place pick points and to run model alignment |
| Point Cloud | A collection of 3D points captured by the camera | The raw depth data used to generate 3D models, to locate objects during execution or handling collisions |
| Pick Point | The point on the object where the robot picks it, with the orientation the tool has to approach it. It can be defined by hand on a 3D model, or generated automatically by Auto-pick or Mask center | It is what Pick[+] turns into the pose sent to the robot |
| Revolution Points | Evenly spaced pick points around a circumference on symmetric parts | Automates point placement for washers, caps, cylinders and rings |
Environment¶
| Term | Definition | Why it matters |
|---|---|---|
| Scene | The representation of the complete work area of the robotic cell, assembled from bins and assets, each placed at its real position relative to a robot base | Defines the workspace Pick[+] uses to compute collisions and to filter detections |
| Bin | The container that holds the objects to pick. It delimits and filters the pick region, and when Solid is enabled its walls also count as a physical obstacle | Restricts the search area on every trigger, and can be selected per cycle by the robot program or the PLC |
| Asset | A obstacle in the cell, such as a table, a structure, a pallet or a fixture | Exists purely for collision handling, so the planned trajectories never intersect it |
| Tool | The gripper or suction cup mounted on the robot flange, described by a geometric model and, for vacuum tools, a suction cup diameter | Sets the TCP every returned pose refers to, and decides which grasps are feasible and collision-free |
| Tool Center Point (TCP) | The reference point on the end-effector from which all robot motions are computed | Ensures consistent, accurate pick and place movements |
| Calibration | Computation of the transform that maps camera coordinates into robot coordinates, either Eye-in-hand (camera mounted on the arm) or Eye-to-hand (camera fixed in the cell) | Guarantees that what the camera sees aligns with where the robot moves. See Camera-Robot Calibration |
Application and AI¶
| Term | Definition | Why it matters |
|---|---|---|
| Application | One complete picking task, defined through a six-step wizard: general data, hardware, collisions, AI strategy, picking configuration and picking strategy | Encapsulates everything needed to run a pick and place task. A single license can hold as many applications as needed |
| AI Model | The neural network that processes the captured image to find the objects in it | Two families exist: CoreVision (detection and segmentation, without classification) and Few-shot (detection, segmentation and classification with embeddings). Both offer the Boxes, Manufacturing and BinPicking variants |
| AI Strategy | How Pick[+] recognizes the objects and where the pick points come from: Smart Picking, Geometry-Based Picking or Similarity-Based Picking | The most important decision of the configuration, since it sets what data the application needs. See AI Strategy |
| Smart Picking | Segments the objects and computes a grasp on each one directly from the captured point cloud | Needs no CAD, no 3D model and no reference, but does not report which object was picked |
| Geometry-Based Picking | Segments the objects and aligns the selected 3D models against the point cloud to recognize them and recover their pose | Gives precise, repeatable grasps on known features. The detection time grows with every model selected |
| Similarity-Based Picking | Segments the objects and classifies them against your enabled references by visual similarity | The only strategy whose AI model classifies what it sees, so every pick is tagged with the reference it was identified as |
| Confidence Threshold | The minimum score a detection needs to become a pick candidate | Raise it to ignore noise and partially visible objects, lower it when valid objects in the bin are being missed |
| Auto-pick | Picking mode that searches the object surface for a flat region wide enough to fit the suction cup diameter | The most tolerant mode when objects are piled and occluded. It requires a vacuum tool |
| Mask center | Picking mode that places one pick point at the center of the object's mask, orienting it from a Principal Component Analysis of its point cloud | Fast and predictable on flat, symmetric or regular parts, with exactly one candidate per object |
| Model alignment | Picking mode that fits a 3D model onto the point cloud to recover the object's pose and transfer its defined pick points onto the real part | Used when the grasp must land on a specific feature, or the part must be picked in a known orientation |
| Candidate | A detected object with a feasible, collision-free pick pose | Several candidates usually exist per capture, and the Picking Strategy decides which one is sent to the robot |
| Pose | A 6-DOF description of a point in space: position X, Y, Z plus orientation RX, RY, RZ |
Specifies the coordinates and orientation the robot has to reach |
Execution and Integration¶
| Term | Definition | Why it matters |
|---|---|---|
| Configuration Mode | Working mode where all the parameters of the software can be defined and modified | Everything on the Home Screen is editable, but the application does not run |
| Execution Mode | Working mode where the active application runs and the Execution Screen is available | Settings are locked, and the application starts automatically when the PC is turned on |
| Trigger | The request that makes the camera capture an image and start a detection | Sent by the application controller, and where the bin, the references and the 3D models can be filtered on each cycle |
| Output request | The request that reads back the computed pick pose and its metadata | Every application needs at least one output receiver, and controller and receiver are usually the same robot |
| Camera Profile | The set of capture settings (exposure, image quality, capture type) applied during a capture | It has to match the object's real working distance for the scan or the detection to come out sharp |
| PLC | External control system that talks to Pick[+] over Modbus TCP or Siemens S7 | It can trigger captures, read results, select which application runs, and address bins, references and 3D models by numeric ID |
| URScript / RAPID module | The robot programming module supplied by Pick[+], the URScript module for Universal Robots and the RAPID module for ABB |
Bridges the Pick[+] server with the robot program through pre-built functions and global variables |
| Category | The name kept in the runtime interfaces for the reference an object was classified as, reported in PP_CATEGORY on a robot and in OUTPUT_CATEGORY_ID on a PLC |
Lets the robot program route each part according to what it turned out to be |
| Digital Output (DO) | Robot I/O line used to control peripherals, such as the LED lighting | Lets the robot trigger lights or other hardware |
| Logs | Runtime records from the client and the server | Crucial for debugging errors or unexpected behavior |