3D CAD Assembly Graphs for Robotic Instruction Generation

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current methods for generating robotic assembly instructions are resource-intensive, consuming significant computing power, processing resources, and time due to iterative programming and data capture processes using augmented reality and motion sensors.

Innovation Solution

A planning system that generates robotic assembly instructions from 3D CAD models using a graph convolution network model to process assembly descriptors, predict component interrelations, and generate assembly instructions efficiently.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If iterative programming and data capture processes using augmented reality and motion sensors are used to generate robotic assembly instructions, then the assembly instructions can be generated with practical accuracy, but significant computing power, processing resources, and time are consumed

Engineering Contradiction:
Improveaccuracy of assembly instructionsVSAvoidcomputing power and processing resources
Core Design Contradiction:
Measurement precisionVSUse of energy by moving object

Solution Approach 1:

The system performs preliminary processing of 3D CAD models to extract geometric features, component relationships, and assembly constraints before generating assembly instructions. By pre-processing the model data and organizing it into structured representations (graphs, matrices), the system eliminates the need for resource-intensive iterative programming and sensor-based data capture during actual assembly operations

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The system creates a virtual digital twin of the physical assembly process by processing 3D CAD models to generate accurate assembly instructions in silico. This virtual modeling approach replaces the need for physical prototyping, augmented reality overlays, and motion sensor data capture, significantly reducing computing and processing resources while maintaining instruction accuracy

Inventive Principle:
Principle #26Copying

2Measurement precision

If iterative programming and data capture processes using augmented reality and motion sensors are used to generate robotic assembly instructions, then the assembly instructions can be generated with practical accuracy, but significant time is consumed

Engineering Contradiction:
Improveaccuracy of assembly instructionsVSAvoidtime to generate assembly instructions
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

The system performs preliminary extraction of assembly semantics, component hierarchies, and geometric constraints from 3D CAD models before instruction generation. This pre-processing step organizes all necessary information into structured data representations, enabling direct generation of accurate assembly instructions without time-consuming iterative programming or sensor-based data capture during operation

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The system creates a complete virtual representation of the assembly process from 3D CAD models, generating accurate assembly instructions through computational geometry processing rather than physical experimentation. This virtual modeling approach eliminates the time required for augmented reality data capture, motion sensor recording, and iterative programming trials

Inventive Principle:
Principle #26Copying

3Use of energy by moving object

If 3D CAD models are processed using a graph convolution network model to generate assembly instructions, then computing resources and processing time are conserved, but the system must accurately predict component interrelations and assembly semantics

Engineering Contradiction:
Improvecomputing power and processing resourcesVSAvoidaccuracy of component interrelation prediction
Core Design Contradiction:
Use of energy by moving objectVSReliability

Solution Approach 1:

The system segments the assembly prediction task into distinct components: extracting geometric features from 3D models, building component graphs representing part relationships, predicting assembly semantics (contact relationships), and generating assembly instructions. This segmentation allows each sub-task to be optimized independently, maintaining high prediction accuracy while minimizing overall computing resource consumption

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The system introduces intermediate data structures (component graphs, adjacency matrices, geometric feature vectors) as mediators between the input 3D CAD models and the final assembly instructions. These intermediaries organize and pre-process information in a way that enables accurate prediction of component interrelations with reduced computational complexity, bridging the gap between raw model data and reliable assembly semantics

Inventive Principle:
Principle #24Intermediary (Mediator)

Data Source

PatentEP4478237B1Generating robotic assembly instructions from three-dimensional computer-aided design models
Publication Date: 2026.01.21 ACCENTURE GLOBAL SOLUTIONS LTD
  • EP4478237B1 patent drawingFigure 1A
  • EP4478237B1 patent drawingFigure 1B
  • EP4478237B1 patent drawingFigure 1C

AI summary

A device may receive a three-dimensional (3D) computer-aided design (CAD) model, and may generate an assembly graph with nodes that represent components and edges that represent contact between the components. The device may generate component graphs for the components, and may generate an assembly descriptor based on the assembly graph and the component graphs. The device may process the assembly descriptor, with a graph convolution network model, to generate node embeddings, and may apply pooling to the node embeddings to generate graph embeddings. The device may calculate a cross attention between the components to generate component interrelations, and may utilize the graph embeddings and the component interrelations to predict links between the components. The device may predict poses and joint axes for the components, and may generate assembly instructions based on the graph embeddings, the component interrelations, the links, the poses, and the joint axes.