Synthetic Data Training for Robotic Depalletizing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for training neural networks to identify boxes for robotic pick and place operations are inefficient due to the need for extensive and costly data labeling, and require significant processing power, which is not feasible in harsh industrial environments.
Innovation Solution
The method employs computer graphics software to generate virtual images of boxes in various orientations and sizes, allowing for automatic labeling and training of neural networks without the need for extensive real-world data collection, using a CPU-compatible approach.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If real-world data collection and manual labeling methods are used to train neural networks, then the neural network can learn accurate box identification, but the time and cost associated with data preparation increases significantly
Solution Approach 1:
The patent uses computer graphics software to generate virtual images that copy real-world box stacking scenarios. These synthetic images serve as training data substitutes, eliminating the need for extensive real-world photo collection and manual labeling while maintaining training effectiveness
Solution Approach 2:
The system automatically generates labeled training data through code-based virtual scene generation. The labeling is performed automatically by the simulation system itself rather than requiring human annotators, making the data preparation process self-service and highly efficient
2Speed
If GPUs are used for neural network processing to achieve faster computation, then processing speed improves, but the system complexity and cost increase, making it unsuitable for harsh industrial environments
Solution Approach 1:
The patent adopts a CPU-based processing approach instead of requiring expensive GPUs. While CPUs are less powerful for neural network computation, they are widely available, more reliable in industrial environments, and sufficient when combined with optimized algorithms and preprocessed data from virtual training
Solution Approach 2:
The system performs extensive neural network training in advance using virtual data generated by computer graphics software. This preliminary training phase allows the model to learn complex patterns offline, reducing the computational burden during real-time industrial operation and enabling simpler hardware to handle inference tasks
3Productivity
If the neural network complexity is reduced by decreasing the number of layers or nodes, then the system becomes more efficient and easier to deploy, but the accuracy of box identification and orientation prediction decreases
Solution Approach 1:
The patent changes the training data parameters by using virtually generated images with precisely controlled variations in box size, orientation, and stacking configurations. This allows the neural network to learn from diverse scenarios without requiring an overly complex architecture, achieving good accuracy with simpler models
Data Source
AI summary
A system and method for training a neural network. The method includes modelling a plurality of different sized objects to generate virtual images of the objects using computer graphics software and generating a placement virtual image by randomly and sequentially selecting the modelled objects and placing the selected modelled objects within a predetermined boundary in a predetermined pattern using the software. The method also includes rendering a virtual image of the placement virtual image based on predetermined data and information using the computer graphics software and generating an annotated virtual image by independently labeling the objects in the rendered virtual image using the software. The method repeats generating a placement virtual image, rendering a virtual image and generating an annotated virtual for a plurality of randomly and sequentially selected modelled objects, and then trains the neural network using the plurality of rendered virtual images and the annotated virtual images.