Very Huge Machine
Exploring how machines could look like if precision relied on intelligence rather than rigidity
My diploma thesis at ABK Stuttgart. The question was whether a machine could learn its own geometry rather than having it precisely engineered.
The machine was a drawing arm built from scrap wood — deliberately imprecise, with no guarantee that a given motor command would move the tool to a predictable position. A camera watched the tool as it moved. Each movement was recorded: motor state in, position out. That data trained a neural network until the machine could produce accurate tool paths despite the looseness of its construction. The drawings were slightly shaky, a product of the wood and the joints. The accuracy came from the network.
