How can we imbue a “maker’s signature” into machine-made textiles?
Procedural Knitting Machine: A Real-time, AI-Enhanced Collaborative Mind-Mapping Tool
Project Details
Role: Software Developer, UX Designer
Skills: Arduino Programming
Project Type: Academic Research Paper
Team: Dominic Co, Amy Chen
Project Date: 2022
Research Paper Link: https://link.springer.com/chapter/10.1007/978-981-19-4472-7_196
Abstract
Once a slow and reflective handcraft embodying a designer/maker’s skill, mood and imaginations, knitting has been automated through computer-aided design(CAD) and manufacturing technologies (CAM), reducing the role of the textile designer in the processes of textile production. In response, Procedural Knit presents a knit procedural generation machine, designed to enable collaboration between the designer/maker and a computational textile design and production technology. The design of the colorwork knitted pattern isn’t predetermined but created through generative design algorithms influenced in real-time by the designer’s posture during the knitting process. Using this system, we explore the design space of “underdetermined fabrication”, an approach where interactive making systems and procedural rules guide but do not determine the final outcome. Through this system, we raised the question of whether the laborious and repetitive nature of producing knit on semi-automated machines could instead be an integral part of the design-through-making process, similar to other handcrafts such as pottery wherein the potter makes active design choices during the pottery’s fabrication. Furthermore, we uncover how interactive underdetermined fabrication can create room for unpredictable emergent phenomena and aesthetics, and notions of imbuing a “maker’s signature” into the resultant knitted artefacts.
Precedent Research
Albaugh (2020) defines Underdetermined fabrication as “interactive systems where a series of procedural rules guide, but do not determine, the final outcomes”. We build upon the experiments of Albaugh et al. (2020) and previous generative textile design works in three specific ways: (1) by exploring untapped modalities of postural input, (2) by exploring the use of a domestic knitting machine as the medium for underdetermined fabrication and (3) by exploring the use of cellular automata to define real-time generative design logic. The postural inputs we focused on provide a source of disruption into the design during the fabrication process itself. These inputs negotiate with the higher level cellular automata rules to guide the process that the user can choose to ignore or follow. The combination of these three directions present, by its own nature, a highly unique and specific set of underdetermined making experiences and design outcomes.
Opportunity: Can we change the knit pattern of a domestic knit machine in real-time as someone is making it?
Challenge: Which positions should we detect and use? And how?
Research-through-Design (Zimmerman et al. 2007) and autobiographical design (Neustaedter, C. and Sengers, 2012) were the methodologies used in this work. These approaches were utilized due to the exploratory nature of the custom knit machine, of which there exists no established ways of using it. In addition, an autobiographical approach enabled the researchers to acquire a more nuanced understanding of system, and to discover significant makes and breaks of the system that might not be otherwise possible with other generalized user-centered approaches; this is due partially to the fact that the researchers-as-users were able to spend more time with the system as opposed to recruited participants. Nippart-eng’s (2015) ethnographic approach of utilizing highly visual and direct observation methods via video-recordings, photographs, and temporal maps were also used to interpret the researcher’s experiences using the system and the highly visual knitted artifacts. Through pre- and post-journaling reflection and videos of our knitting process to cross-check data, our engagement in the planning, implementation, and use of the Procedural Knit system served as an opportunity to more deeply understand the limitations and considerations of designing and using underdetermined fabrication systems. Given that the systems engage the user’s entire body in highly physical ways, we viewed these systems as personal design systems that need to be changed based on preferences.
Through video recordings and direct observation, we recorded what kind of affordances we had when using the Toyota KS585, a single bed domestic knitting machine. It was observed that the knitting experience involved a variety of steps and movements. These include pushing and pulling the carriage [A], slouching overtime [E] and rotating on the swivel chair the user is seated on [L] to name a few (below). We ideated on the possibility that our knitter’s movements and experiences could be tangibly linked to the design of the knit itself as an underdetermined fabrication process. In this project, we focused on slouching [E] and rotating movements [L].
Cellular Automata Implementation
A cellular automaton is a computational model that consists of a grid of squares which can have one of two states, ‘on’ or ‘off’. Cell states are determined by the state of their neighbouring cells and a set of rules (Shiffman, 2012). Out of the 256 possible CA rules, Rule 30 was of particular interest due to the seemingly complex, yet visually legible pattern it generates.
Due to the knitting machine’s 12 needle patterning limit, corresponding to the 12 cells in the CA algorithm, the CA algorithm generates a pattern that eventually transforms into a monotonous, regular striped pattern after around the 40th row irregardless of the seed row (below). To address this, a pseudo left-hand-side (LHB) and right-hand-side “buffer” (RHB) cell were added, making the total number of 14 cells per row. These buffer cells were not directly linked to the 12 button pattern instructions the knitter can follow, but rather served as containers for data recorded from the knitter’s postural inputs, which disrupt the pattern generated. Changes in the buffer cells added variation into the pattern as a result of changes in posture. This worked in adding moments of disruption i.e. complex CA patterns which counter-balanced moments of monotony presented by the striping pattern (Figure 4). In our experiments, we started of with a seed row of alternating 1s, and 0s that produced the regular stripped pattern. This was done in order to visualize the effect of postural inputs would have on the generated CA pattern in our system.
Experiments and Results
In two primary experiments, we developed and evaluated slouch and rotation detection systems to capture and reflect a knitter’s posture and movements during the completion of 400 rows over two hours. The slouch detection system initially monitored posture changes, resulting in altered knit patterns and causing the knitter significant mental and physical exhaustion. In its second iteration, incorporating a harness effectively reduced slouching and produced a consistent striped pattern, though it introduced unique pattern errors (white streaks) that suggested the knitter’s mental state was imprinted in the design. Similarly, the rotation detection system first mapped the knitter’s rotational movements to pattern variations, and its subsequent iteration utilized striped patterns to more clearly visualize these turns. This revealed that the knitter’s position relative to the machine influenced pattern activation. Overall, the experiments demonstrated that tracking posture and movement can effectively embed the knitter’s physical and mental states into the resulting knit designs.
