From imprecision to precision in virtual reality: a machine learning approach to parametric CAD modelling

This paper introduces a Machine Learning (ML) method for translating Virtual Reality (VR) based interactions into a sequence of Computer Aided Design (CAD) operations compatible with typical CAD toolchains. The method involved creating an integrated toolchain of ML, VR, and a CAD kernel, and uses VR-based sculpting as an exemplar design tool. This tool is used to curate a dataset of imperfect voxel-based representations of CAD features, which then augments the existing perfect CAD model data to train an ML model to recognise user interactions as the intended design features. A total of eight design features were created that can generate 24 different manufacturing features. The implemented system showed a validation accuracy of 95.6%, a test accuracy of 84.2%, and a Top 3 test accuracy of 97.1%. As such, the work demonstrates the viability of ML-augmented spatial toolchains for creating parametric CAD geometries and can pave the way for a new direction in developing VR-CAD modelling software.