Robotic Work Positioning Using Synthetic Instance Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing machine learning models for instance segmentation, such as Mask R-CNN, require extensive manual data preparation and are not optimized for real-world engineering applications, leading to inefficiencies in determining accurate work positions for robotic systems handling bulk-stacked objects.
Innovation Solution
A system that uses a machine learning device to automatically generate and learn instance segmentation models by arranging virtual objects in a virtual space, generating virtual object images, and training models using these images to identify work regions and positions, thereby reducing the need for manual data creation and improving computation speed.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual data preparation is used for training instance segmentation models, then model accuracy can be improved, but the time and labor required for data creation increases significantly
Solution Approach 1:
The patent uses synthetic image generation to create virtual copies of objects and scenes, replacing the need for manual photographing and annotation of real objects. The system generates training data by rendering 3D models in various virtual environments, which provides unlimited diverse training samples without requiring physical data collection and manual labeling efforts.
Solution Approach 2:
The system performs preliminary data preparation by pre-rendering synthetic training images and annotations before actual model training begins. This includes pre-processing 3D models, generating multiple views and variations, and creating complete annotation sets in advance, thereby eliminating the need for time-consuming manual data preparation during the model development phase.
2Reliability
If extensive manual data preparation is performed, then training data quality improves, but the complexity and cost of the data creation process increases
Solution Approach 1:
The patent replaces manual mechanical processes of photographing, physically manipulating objects, and hand-annotating images with automated computer-based synthetic image generation. The system uses algorithmic rendering engines to automatically generate images and annotations from 3D models, eliminating the need for human operators and complex manual data pipelines.
Solution Approach 2:
The synthetic image generation system is self-sufficient, automatically generating both images and their corresponding annotations without external intervention. The system autonomously renders images from 3D models, detects object boundaries, generates segmentation masks, and creates bounding box annotations programmatically, making the data creation process independent of manual labor.
3Measurement precision
If traditional object detection methods like Faster R-CNN are used, then object existence regions can be obtained, but computation speed decreases due to processing all pixels
Solution Approach 1:
The patent extracts and processes only the regions containing objects of interest rather than analyzing all pixels in the image. By using synthetic training data that emphasizes object boundaries and regions, the model learns to focus computational resources on relevant areas, effectively extracting object regions without exhaustive full-image processing.
Solution Approach 2:
The system implements instance segmentation that divides the image into distinct object regions, processing each object instance separately rather than treating the entire image as a single processing unit. This segmentation approach allows the model to identify and process only the necessary regions containing objects, improving computational efficiency while maintaining detection accuracy.
Data Source
AI summary
Provided is a work system including: an object imaging unit configured to acquire an object image by photographing an object from a work direction; a work position acquisition unit configured to acquire a work position based on an existence region of the object obtained from a machine learning model; and a work unit configured to execute work on the object based on a work position obtained by inputting the object image to the work position acquisition unit.