In recent years, Artificial Intelligence (AI) has seen some incredible progress at completing tasks that were thought to be only possible by humans such as speech recognition, sentiment analysis, and even producing visual art. However, AI models still struggle to capture complex tasks that are constrained by different human and technical parameters. One example of such a task is Computer-Aided Design (CAD) that is a technical process driven by human intuition. It can take skilled engineers years to master CAD modelling as it requires learning a sound combination of design and technical skills. By gaining experience, the engineers develop a design intuition which allows them to think ahead and take the right actions at all stages of the design process.
CASCADES will use cutting-edge AI technology to make machines learn this design intent. In particular, technical constraints will be used as clues to uncover the thought processes guiding CAD designers. Such technology could be directly applied to reverse engineer the CAD construction history of a physical object, thus allowing to save resources by automating industrial design. The interest in this reverse engineering process has been encouraged by recent advances in 3D handheld scanner technology.
As a result, the ultimate goal of CASCADES is to develop a complete pipeline to automate the 3D reverse engineering of CAD models from 3D scans by mimicking the design intent. CASCADES will be conducted in close collaboration with one of the world leaders in the field of 3D technologies based in Luxembourg, Artec3D. By automatically learning the design intent, it is hoped that CASCADES will bring us one step closer towards modelling the human mind using machines.
- Starting date: 01/04/2022
- Duration: 48 months
- Funding source: FNR Industrial Fellowships
- Researchers: Elona Dupont (PI), Sk Aziz Ali, Dr. Anis Kacem, Prof. Djamila Aouada
- Partners: Artec3D
Elona Dupont, Kseniya Cherenkova, Anis Kacem, Sk Aziz Ali, Ilya Arzhannikov, Gleb Gusev, and Djamila Aouada, “CADOps-Net: Jointly Learning CAD Operation Types and Steps from Boundary-Representations.” In IEEE International Conference of 3D Vision (3DV), 2022. [Project Page]