Topic: Exist Machine Learning or Artifitial Intelligence for Autocad?

[It's Alive!]

not possible with all the crazy architects and designers 

Ahhh sorry but it already here.  In the form of Google Sketch Up.   

And it should not be a surprise that Google was the one to bring it.

Print!  “I’m sorry Dave, I’m afraid I can’t do that”

[Mark]

not possible with all the crazy architects and designers 

Ahhh sorry but it already here.  In the form of Google Sketch Up.   

And it should not be a surprise that Google was the one to bring it.

Print!  “I’m sorry Dave, I’m afraid I can’t do that”

LOL literally.

[jtm2020hyo]

....just curious... why BricsCAD and not AUTOCAD?
can I do it with an OPEN SOURCE alternative?

I mentioned BricsCAD because it's a cheaper alternative to AutoCAD, and your post made it sound like you wanted to implement it as a personal project. You could certainly use AutoCAD.

As for an open source alternative, I see no reason you couldn't do it. Heck, it might even be easier if the open source CAD alternative you're looking at has native python support. Many of the machine learning libraries in python are open source.

After my first post, I felt like I wasted time responding to your question. I may have misinterpreted, but it almost seems like you're looking for a turn-key answer for you to use (free of charge no less).

As far as I know (and I've been looking) that answer doesn't exist. It's gonna take work. Lots of it. Work worth doing IMO, but work none the less.

From your original post:

I found some videos where explain that   Machine Learning and Artificial intelligence can be used for everything and I can teach to the Machine or A.I. to draw like my old projects.

Whenever a statement (even on the internet ) applies it's self to everything chances are it's probably a false statement.

If anyone in this community wanted to start up a team doing some of this work, I would be happy to contribute some time and attention to it.

you did not waste time. I just searching for a "TEMPLATE" to start my project... at least that as I always start to learn anything.

[kdub_nz]

jtm2018hyo,

Which languages do you program in ??

[jtm2020hyo]

from: kdub on 2019-04-04, 17:17:03
jtm2018hyo,

Which languages do you program in ??

I just programmed a beautiful and perfect AUTOEXE.cfg for bind keys in DOTA 2, and a beautiful and perfect AUTOhotkeys for my Warcraft III.

basically, I'm like Stev Jobs ... but programming a bit.

[kdub_nz]

I understand ...
you like playing games.

[kdub_nz]

I've been putting some thought to this and can't determine any areas of CAD production that would benefit from Machine Learning and Artifitial Intelligence.


I know, I don't get out very much
Anyone have any realistic ideas ??

... even without the proviso of being financially viable.

Regards,
       

[MP]

I've been putting some thought to this and can't determine any areas of CAD production that would benefit from Machine Learning and Artifitial Intelligence.

That won't stop the overlords.

[Atook]

...
Anyone have any realistic ideas ??

kdub, I don't think it will help CAD production for a while. Right now it's about design. And really about design that has multiple good solutions; flexibility if you will. If you can code the design requirements into a rigid algorithm with all constraints known and one single solution, you don't need machine learning, just run the algorithm. It'll be way faster, cheaper and efficient.

If you have a problem with multiple good solutions, that's where the magic of ML will come in. So if the algorithm is 'loose' machine learning will run thousands/millions of iterations, while optimizing certain outputs. It comes up with stuff that doesn't make logical sense, but work.  Some of the solutions will suck, and some will be really good, and appear to be creative because designers tend to limit themselves to what they know works.

The cool thing I think is, it would be possible to 'teach' the machine learning algorithm using a bunch of previous designs done by a human.  At that point, it might be possible for ML to come up with designs that pull from previous designers experience, but still iterate and generate new solutions. Or once the ML algorithm is tuned, it'll just be faster. And if we can figure out how to use the thousands of existing  drawings as a learning dataset, all the better.

My understanding is ML is being used to design airplane parts now. And I'd love to see it expand into other areas. I have ideas on how to apply it to my current area of design (landscape irrigation) but I haven't figured out the nature of the implementation. As I said in a previous post, just figuring out how to ask the question correctly is difficult.

Regarding AI, as far as I know it's not here yet, and would be a waste of time trying to develop it for CAD. Once/if AI shows up, all bets are off. It's not worth worrying about IMO. Anyone claiming to have achieved AI hasn't done a good job of reigning in their marketing dept.

[kdub_nz]

>>>>   And if we can figure out how to use the thousands of existing  drawings as a learning dataset, all the better.    <<<

Just make sure the drawings have the lines joining up at the corners.

but seriously;
I may be naive and a bit of a romantic but I can't visualise 'design' without input of a human brain ... without a brain and emotion it is just sausage making.

Current software evaluates multi-thousands of emails trying to learn to identify fraud and spam ... and that's with a major percentage of the 'rules' being written by human hands. It's a big step from there to designing a bridge ( or an irrigation system ).

Wearing a different hat:
An irrigation system would be feasable ( without ML and AI ) once the rules were established and a full reference database was created. I don't think you'd find it too difficult, just time consuming.

Regards,

[dgorsman]

Most systems design has rules, however in some cases the rules can get so complex that a linear calculated solution isn't very practical.

The first example would be laying out pipes through a pipe rack.  You would have source and destination equipment or other piping; commodities and properties; etc.  A good designer could lay out a design in a couple of days, while an AI could iterate through hundreds of potential designs within an hour, and cull out the top performers for further development.

The second example is a little more drafting oriented - automatic tag layout.  Again, there are all sorts of rules - don't cross the streams (er... leaders  ); don't overlay annotations; keep annotations close to the point where they are annotating; etc.  You could do some rough calculating but the end results wouldn't be pretty, and would likely fail on those drawings which don't fall within certain strict parameters.  On the other hand, an AI could evaluate the potential options and find a top-performer that is "good enough" using a genetic algorithm (which is something I've got on the go).

[kdub_nz]

I haven't looked further than the release notice, but this may be better than re-learning R and Python with all it's add-ons

https://visualstudiomagazine.com/articles/2019/04/08/mlnet-rc.aspx

or not, YMMV

[kdub_nz]

http://www.theswamp.org/index.php?topic=55115.0

[jtm2020hyo]

hello, I have some question:

about Autodesk licenses, can be AutoCAD can be used to get profit for (example) myself?

I mean, in the hypothetical case that I get some advances in Machine Learning to automate my work in AutoCAD, Autodesk could you ask me for some money?

if the answer is yes, then have no sense work with Autodesk software, then, what software might suggest to me?

PD: sorry for the late answers, this is just a hobby and I will just join in my free time, I would never leave it in oblivion.

[Atook]

Autodesk would only ask you for the subscription fee you agreed to.

I could see that they might make an AutoCAD version with a built in ML module that would cost more. But I think that's long in the future.

Drafting and design is a tough set of problems for machine learning so far. The questions aren't as well defined as Hot Dog or not Hot Dog