Robot-Guided Camera Training for Automatic Object Pose Recognition

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing machine learning methods for object recognition in robotics require manual marking of rectangular areas in numerous training images, especially when handling different objects, which is time-consuming and inefficient.

Innovation Solution

A method involving positioning a robot-guided camera in various poses to capture localization images, determining a virtual model of the learning object, and using machine learning to automate the determination of object poses, particularly through deep artificial neural networks, allowing for flexible and reliable object recognition.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If manual marking of rectangular areas in training images is used, then object recognition can be learned, but the process is time-consuming and inefficient

Engineering Contradiction:
Improvetraining efficiencyVSAvoidmanual marking time
Core Design Contradiction:
ProductivityVSLoss of time

Solution Approach 1:

The system performs automatic pose determination and virtual model generation without requiring manual marking. The robot-guided camera system captures images, determines poses automatically using the virtual model, and generates training data self-service style, eliminating the need for human annotators to manually mark rectangular areas in training images

Inventive Principle:
Principle #25Self-service

Solution Approach 2:

The manual mechanical process of marking rectangular areas is replaced by an automated computational system. The system uses virtual models, pose determination algorithms, and machine learning to automatically generate training data, substituting the manual marking mechanism with an automated digital process

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

2Extent of automation

If a robot-guided camera is used to capture localization images from different poses, then automated object recognition can be achieved, but the system complexity increases

Engineering Contradiction:
Improveautomation levelVSAvoidsystem complexity
Core Design Contradiction:
Extent of automationVSDevice complexity

Solution Approach 1:

The robot system serves multiple functions: it positions the camera to capture images, determines poses of objects, generates virtual models, and provides training data for machine learning. This multi-functional approach consolidates what could be separate systems into one unified platform, managing complexity through functional integration

Inventive Principle:
Principle #6Universality (Multi-functionality)

Solution Approach 2:

The virtual model acts as an intermediary between the captured images and the pose determination process. Rather than directly analyzing complex images to determine poses, the system uses the simplified virtual model as a mediator to facilitate automated pose determination and training data generation

Inventive Principle:
Principle #24Intermediary (Mediator)

3Measurement precision

If deep artificial neural networks are used for machine learning, then object recognition accuracy improves, but computational requirements and processing time increase

Engineering Contradiction:
Improverecognition accuracyVSAvoidcomputational energy
Core Design Contradiction:
Measurement precisionVSUse of energy by moving object

Solution Approach 1:

The system performs preliminary actions by automatically generating high-quality training data with precise pose information using the robot-guided camera and virtual model approach. By preparing accurate training data in advance, the neural network can achieve high recognition accuracy more efficiently during the actual object recognition task, reducing the computational energy needed for inference

Inventive Principle:
Principle #10Preliminary action

Data Source

PatentEP3966731B1Machine learning of object recognition using a robot-guided camera
Publication Date: 2026.01.28 KUKA DEUT GMBH
  • EP3966731B1 patent drawingFigure 1~2

AI summary

The invention relates to a method for machine learning an object detection process using at least one robot-guided camera (12) and at least one learning object (30), having the steps of: - positioning (S10) the camera in different positions relative to the learning object using a robot (10), wherein at least one localization image, in particular a two-dimensional and/or a three-dimensional localization image, which images the learning object, is captured in each position and is stored; - ascertaining (S20-S60) a virtual model of the learning object on the basis of the positions and at least some of the localization images; - ascertaining (S70-S90) the position of a reference of the learning object in at least one training image captured by the camera, in particular at least one of the localization images and/or at least one image with at least one interference object which is not imaged in at least one of the localization images, on the basis of the virtual model; and - machine learning (S100) an object detection of the reference on the basis of the ascertained position in the at least one training image. A method according to the invention for operating a robot (10) has the steps of: - ascertaining (S130, S140) a position of at least one reference of an operating object (30') using the object detection process which has been learned using the method; and - operating (S150, S160) the robot on the basis of said position.