carveRL
Serendipity: the chance of making positive discoveries is much greater in a craftsperson's environment than in an industrial fabrication setting. But with the rise of digital fabrication, we are losing the value of those fortuitous mistakes.
carveRL integrates machine learning into the fabrication process to bring serendipity back. Instead of programming a tool path, the machine is taught to operate within its environment, free to find those accidental discoveries that lead somewhere interesting.
The case study: pottery. Drawing from Wedgwood's lathe and the Mishima carving technique, carveRL is a tool trained to carve patterns into clay, and keep surprising its maker.
Ver 1
A linear actuator learns to follow a specific pattern. The agent avoids cylinders and white balls while making contact with the black ones, and its physical twin replicates each movement in real space simultaneously.
Ver 2
A two-axis carving tool trained in Unity3D ML Agents. Multiple tools learn to avoid cylinders and white balls while seeking red ones, without any prescribed path. Once trained, rearranging the balls produces entirely different patterns.
Collaboration: Yael Rosenberg
To test the tool's ability to collaborate with a ceramic artist, Yael Rosenberg created patterns by hand, then programmed the machine to attempt similar ones. Throughout the process, serendipity occurred, and Yael discovered new directions she hadn't planned.
Results
A selection of patterns produced using the two-axis carving tool.
