Augmented Carpentry: Computer Vision-assisted Framework for Manual Fabrication
Andrea Settimi, Julien Gamerro, Yves Weinand
TL;DR
Augmented Carpentry (AC) introduces a computer vision–assisted, open-source framework that retrofits ordinary woodworking tools with sensing and AR guidance to support manual timber fabrication. It combines toolhead pose estimation with a timber-centric SLAM (T-SLAM) and a digitally locked execution model to provide real-time overlays and a 3D recording of fabrication decisions, enabling human–computer collaboration without rigid workflow enforcement. In a 1:1-scale experimental campaign (166 joints across 57 spruce beams), AC achieved mean joint-position errors below 3 mm, sub-millimeter joint-face accuracy, and drilling angles around 1.2°, with some degradation for beams longer than ~3 m due to map quality. The work demonstrates a path toward democratizing digital timber fabrication by retrofitting existing tools, reducing reliance on high-end robotics, and enabling traceable, on-site fabrication workflows that integrate human skill with digital guidance.
Abstract
Ordinary electric woodworking tools are integrated into a multiple-object-aware augmented framework to assist operators in fabrication tasks. This study presents an advanced evaluation of the developed open-source fabrication software Augmented Carpentry (AC), focusing on the technical challenges, potential bottlenecks, and precision of the proposed system, which is designed to recognize both objects and tools. In the workflow, computer vision tools and sensors implement inside-out tracking techniques for the retrofitting tools. This method enables operators to perform precise saw-cutting and drilling tasks using computer-generated feedback. In the design and manufacturing process pipeline, manual fabrication tasks are performed directly from the computer-aided design environment, as computer numerical control machines are widely used in the timber construction industry. Traditional non-digital methods employing execution drawings, markings, and jigs can now be replaced, and manual labor can be directly integrated into the digital value chain. First, this paper introduces the developed methodology and explains its devices and functional phases in detail. Second, the fabrication methodology is evaluated by experimentally scanning the produced one-to-one scale mock-up elements and comparing the discrepancies with their respective three-dimensional execution models. Finally, improvements and limitations in the tool-aware fabrication process, as well as the potential impact of AC in the digital timber fabrication landscape, are discussed.