Horizontal-Surface Item Picking With Surprisal-Based 3D Vision
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing automated systems for manipulating box-like items face challenges in adaptability, efficiency, and reliability due to reliance on classical image processing or machine learning, which are time-consuming, expensive, and require extensive retraining for new items, while manual processes are repetitive and prone to injury.
Innovation Solution
A method and system using a 3D or depth sensor with multipoint disparity technology to generate and rank hypotheses based on surprisals, enabling efficient and adaptable manipulation of multiple items on a horizontal support surface, employing a vision-guided robot for precise handling.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Extent of automation
If classical image processing or machine learning is used for automated item manipulation, then automation is achieved, but the system requires extensive retraining and is time-consuming for new items
Solution Approach 1:
The patent replaces traditional mechanical vision systems (classical image processing) and complex machine learning models with a physics-based light transport model. This analytical model calculates how light travels from illumination sources to camera sensors, enabling the system to understand item geometry and configuration without training. The substitution of data-driven approaches with physics-based reasoning achieves both automation and immediate adaptability to new items.
2Adaptability or versatility
If manual processes are used for item picking, then flexibility is maintained, but human labor is repetitive and prone to injury
Solution Approach 1:
The system enables automated manipulation where the robotic system serves itself by using the physics-based model to automatically understand item configurations, plan manipulation sequences, and execute tasks without human intervention. This self-service capability eliminates repetitive manual labor while maintaining the flexibility that humans previously provided, thereby improving safety and reliability.
3Measurement precision
If extensive training and reconfiguration are implemented for new items, then accuracy is improved, but time and cost increase significantly
Solution Approach 1:
The patent performs preliminary action by pre-computing the physics-based light transport model for the camera and illumination setup. This pre-computed model can then be applied immediately to any new item configuration without retraining or reconfiguration. The preliminary establishment of the analytical framework enables both high accuracy and immediate adaptability, eliminating time losses associated with training.
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
The system provides efficient, adaptable, and reliable automation of box picking and decanting processes, reducing human labor risks and improving manufacturing efficiency without the need for extensive training or reconfiguration.
Implementation Method 1
A method and system using a 3D or depth sensor with multipoint disparity technology to generate and rank hypotheses based on surprisals
Data Source
AI summary
A method and system for manipulating (i.e. such as by picking) a multitude of target items supported on a substantially horizontal support surface one at a time. The support surface supports a configuration of substantially identical items aligned substantially perpendicular to the support surface. The method includes the steps of providing a plurality of hypotheses which are ranked based on surprisals of the hypotheses. Each of the hypotheses describes an observation of an item in the configuration. The observations include an observation of the appearance of a perimeter of the item and an observation of the geometry of the perimeter of the item. The method also includes generating potential configurations for potential combinations of the multiple items based on the ranked hypotheses. Finally, the potential configurations are ranked. The step of ranking includes the step of combining the surprisals of the hypotheses in each potential configuration.