Distributed dataflow machine controllers

Abstract

Workflows in Digital Fabrication require coordination across heterogenous computing systems, from the design tools used to describe component geometries to the embedded control systems used to interact with the physical world in order to produce those components. In the state of the art, workflows are typically static and opaque, especially within embedded controllers themselves. This makes them difficult to modify or develop, and places barriers between high level computing and low level control. An opportunity exists to develop an open platform for interoperability and reconfigurability that spans low- and high-level workflow components, that could collapse much of the heterogeneity found in these systems into cohesive representations. To do so, this thesis develops a systems architecture based on reconfigurable graphs of dataflow objects. It embeds virtual dataflow graphs of modular software elements within physical dataflow graphs of modular hardware elements, recasting heterogenous systems as cohesive graphs all the way down. The architecture is reduced to practice across high-level browser computing and low-level embedded control, through mixed networking links. It is deployed on two machine systems: one that collapses path planning and path execution for a small milling machine, marking a departure from the historic use of G Codes, and another that aligns computer vision based measurement with low-level motor control and sensor acquisition, to open access to materials measurement.