Learning computer-aided manufacturing from demonstration: a case study with probabilistic movement primitives in...

Area of Science:

• Robotics
• Digital Fabrication
• Human-Computer Interaction

Background:

• Conventional Computer-Aided Manufacturing (CAM) tools are limited to standard processes, hindering craftspeople using non-standard operations.
• Formalizing tacit knowledge for custom CAM routines is challenging for makers and craftspeople.
• A gap exists in adapting CAM for unique, craft-like manufacturing tasks.

Purpose of the Study:

• To integrate Learning from Demonstration (LfD) into digital fabrication workflows.
• To enable makers to establish new manufacturing operations through manual demonstrations.
• To develop a CAM pipeline for craft-like operations using LfD.

Main Methods:

• A case study on robot wood carving using hand tools.
• Integration of probabilistic movement primitives (ProMPs) within Rhino's Grasshopper environment.
• Recording human demonstrations via kinesthetic teaching and modeling with ProMPs.

Main Results:

• A pipeline was developed to translate 2D drawings into skill-informed 6-DOF carving toolpaths.
• The system functions as a translator from drawing input to toolpath output within a CAD environment.
• Demonstrated on multiple wood carving applications.

Conclusions:

• Learning from Demonstration (LfD) shows potential for augmenting CAM tools for specialized manufacturing.
• The developed pipeline allows makers to generate custom toolpaths from simple drawings.
• Further research is needed on representing carving skills for generalized toolpath generation.