Is it only a matter of time before you can get generative AI to create a pattern based on a prompt and then some service mails gift wraps and mails it to you/friend (along with video of the lego machine making it)? Just in time for the holiday season :D
Added: there are lots of internet discussions about how crochet, unlike knitting, cannot be automated. But deeper digging turns up a company called Comez and a company called Taiwan Dahu that seem to make massive industrial ones.
I have a vision of a device kind of like a pair of scissors, except instead of shears there are two levered attachments: a holder for a needle and a hook to manipulate yarn.
Every time you close and open the handles, it puts one standard knot onto the needle. This could make knitting a row as easy as snip snip snip, but I don't know if it would actually be useful in knitting real projects.
I've been curious about knitting & crochet from a programming / robotics / simulation perspective for a while, but deferred it to cut my teeth on simpler problems. Those industrial robots seems to be knitting, not crocheting. They use the term "crochet knitting" as a sales tactic.
This short video is annoying but informative: [1]
The closest I've seen to actual robotic crochet is [2].
Which is exciting and close - but it's hard to overstate its limitations. I took the challenge in that short video seriously and spent 10-20 hours learning basic crochet. It became very clear that replicating my simple test patterns would require vision, planning and modeling capabilities beyond anything I've seen in SOTA surgical robots.
What I find interesting about the post (Knotty) is buried here [3] - apparently it's possible to ditch the grid and create an intuitive representation of the final knit pattern. I suspect that may be doable using traditional algorithms.
You have to admire the full-circle of history about this. Arguably, the first stored programs were created in the 1820s to automate Jacquard looms. The looms used a series of punch cards to automate the weaving of complex patterns. The system helped inspire Charles Babbage's Analytical Engine.
Now 200 years later, we get Knotty. Very satisfying, in a holistic interconnectedness sort of way.
As a knitter, I feel compelled to point out that weaving and knitting are very different ways to create textiles. (And crochet is even more different, which is why there are currently no machines that can crochet.)
Like many an invention, he was standing on the shoulders of others
Basile Bouchon (1725) created one of the earliest automated looms using perforated paper tape to control the weaving pattern.
Jean-Baptiste Falcon (1728) improved on Bouchon's design by using chains of punched cards instead of continuous paper tape, making the system more durable and easier to handle.
Jacques Vaucanson (1745) further refined the concept with his own punch card loom design.
Jacquard's breakthrough came in 1804-1805 when he synthesized and perfected these earlier innovations into what became known as the Jacquard loom. His version was more reliable, easier to operate, and became widely adopted - which is why his name became so strongly associated with the technology.
Thanks, I learned something today. If Wikipedia is to be believed:
> He played an important role in the development of the earliest programmable loom (the "Jacquard loom"), which in turn played an important role in the development of other programmable machines, such as an early version of digital compiler used by IBM to develop the modern day computer.
There are examples of these in the London Science Museum. I strongly recommend a visit - and not just for those: there's also a (working) full-scale replica of Babbage's difference engine; the entire early history of steam engines (the genuine articles); a duplicating lathe, from the early 19th century; models of the entire history of the (British, at least) tractor; an entire analogue telephone exchange; and way, way, way more. It's a geek's delight. One of my favorite museums ever, but be warned: it took me more than a week's worth of visits to feel like I'd seen it thoroughly.
I'd still like to see some 2-seams-only T-shirts as should be straight-forward on a sufficiently-flexible tube knitting machine: seams down along the bottom of the arms continuing to the hips, so there's no seam around the shoulders.
Just double up columns appropriately; maybe the seam doesn't even have to go all the way down.
But that column count is out of reach for home style machines.
24 comments
[ 2.9 ms ] story [ 40.8 ms ] threadIs it only a matter of time before you can get generative AI to create a pattern based on a prompt and then some service mails gift wraps and mails it to you/friend (along with video of the lego machine making it)? Just in time for the holiday season :D
Added: there are lots of internet discussions about how crochet, unlike knitting, cannot be automated. But deeper digging turns up a company called Comez and a company called Taiwan Dahu that seem to make massive industrial ones.
Every time you close and open the handles, it puts one standard knot onto the needle. This could make knitting a row as easy as snip snip snip, but I don't know if it would actually be useful in knitting real projects.
This short video is annoying but informative: [1]
The closest I've seen to actual robotic crochet is [2].
Which is exciting and close - but it's hard to overstate its limitations. I took the challenge in that short video seriously and spent 10-20 hours learning basic crochet. It became very clear that replicating my simple test patterns would require vision, planning and modeling capabilities beyond anything I've seen in SOTA surgical robots.
What I find interesting about the post (Knotty) is buried here [3] - apparently it's possible to ditch the grid and create an intuitive representation of the final knit pattern. I suspect that may be doable using traditional algorithms.
[1] https://www.youtube.com/watch?v=EImnSsCadK8
[2] https://www.youtube.com/watch?v=T1-pfeaVsOM
[3] https://stitch-maps.com/about/overview/
Now 200 years later, we get Knotty. Very satisfying, in a holistic interconnectedness sort of way.
https://en.wikipedia.org/wiki/Jacquard_machine
Basile Bouchon (1725) created one of the earliest automated looms using perforated paper tape to control the weaving pattern.
Jean-Baptiste Falcon (1728) improved on Bouchon's design by using chains of punched cards instead of continuous paper tape, making the system more durable and easier to handle.
Jacques Vaucanson (1745) further refined the concept with his own punch card loom design.
Jacquard's breakthrough came in 1804-1805 when he synthesized and perfected these earlier innovations into what became known as the Jacquard loom. His version was more reliable, easier to operate, and became widely adopted - which is why his name became so strongly associated with the technology.
> He played an important role in the development of the earliest programmable loom (the "Jacquard loom"), which in turn played an important role in the development of other programmable machines, such as an early version of digital compiler used by IBM to develop the modern day computer.
[0] https://en.wikipedia.org/wiki/Joseph_Marie_Jacquard
For i = 1:20
Physical Reality as Type System
- Knitting notation treats physical constraints (yarn direction, needle capacity) as compile-time invariants, not runtime checks
- The notation literally cannot express impossible operations - unlike most programming languages where invalid states are runtime errors
Semantic 2D Syntax
- Spatial positioning encodes operational meaning, not just formatting
- Chart coordinates directly map to fabric coordinates - position IS semantics
- Most programming languages waste the 2D plane on purely aesthetic layout
Context-Dependency That Works
- Heavy context-switching (RS/WS, cable positions) remains manageable because contexts map to physical realities practitioners can feel
- Suggests context-dependent syntax works when contexts correspond to user’s embodied experience
- Contradicts typical CS wisdom that context-dependency is always bad design
Expert Chunking Over Beginner Clarity
- Single symbols encode 4+ sequential operations for expert efficiency
- Notation evolved to optimize pattern recognition for power users, not learning curves
- Inverse of typical programming language priorities (readability over expert speed)
Domain Invariants Enable Multi-Modal Translation
- Knotty’s translation capabilities required modeling yarn physics, not just syntax
- The type system encodes gauge, tension, and material properties as first-class constraints
- Shows DSL success requires modeling domain physics, not just domain vocabulary
Power-Law Feature Adoption
- 10 symbols for basics, 200+ for experts - mirrors programming language feature creep
- But knitting notation maintained backwards compatibility across complexity levels
- Suggests sustainable complexity growth patterns for language design
Just double up columns appropriately; maybe the seam doesn't even have to go all the way down.
But that column count is out of reach for home style machines.