So, I’ve been using Fusion 360 to design a new router base plate system for my trusty old Bosch 1613EVS router (hey, made in USA and has great depth micro-adjust). And I always struggle with which contour/pocket toolpaths to use: 2D vs 3D vs adaptive, etc.
Which got me thinking - with AI in everything these days, why doesn’t AutoCAD have an AI-generated toolpath feature? Seriously:
Pick a body or component to Manufacture.
Tell it what machine you have (already done in Setup).
Tell it what bits have have (rather than choosing ones for specific toolpaths, just tell it all the bits you have available. You only need to list them once).
Tell it the material (already done), how the stock is secured (already in Setup), etc.
Give me the Toolpaths!
Really now, it should be this simple. If AutoCAD would do this, I’d actually pay for a subscription.
And Carbide3D, if you’re reading, this would be great in Carbide Create, too.
Yep, I’m pretty sure Saunders from Saunders Machine Works has mentioned an AI tool path generation company in their workflow. Not sure if it’s the same one but apparently it does a pretty good for getting the bulk of the work done, then an experienced operator does the work to clean up and verify.
It’s not “Fully” automated. There are several steps. Recognize features, Create feature groups, Create toolpaths. But each of those steps are highly automated.
The other caveat is, the system is set up for customers to input their own machining processes. It comes preloaded with sample data, but most customers want to customize it to match their tried & true processes. It’s a pretty big effort to get it all set up.
It also comes preloaded with automatic machining data that sets feedrates, depth of cut, stepover etc… But it’s set up for commercial machines mostly cutting metal, and doesn’t include materials like wood, plastics, composites. It also isn’t set up for these type of machines (high speed spindle, router tooling, low to moderate rigidity. Those could all be added, but again quite an effort.
The last caveat is the cost. It varies quite a bit with many options, but even a basic “essentials” CAD with fixed axis machining is going to be 5 digits.
I would not trust AI toolpaths for a long time. I am a Software Engineer and have had to bring AI tools into my daily workflow. While they are great at generating 90% of what you need done, they are confidently wrong about that last 10%. I end up spending 50% of my time debugging the weird bugs the AI code generated or having to provide the AI feed back as to what it did wrong and get it to regenerate the code. On top of that the code is far from optimized. It tends to generate algorithms that a first year computer programmer would make.
To generate AI toolpaths properly and above all safely you would need to have a feedback system where the toolpath was checked by some sort of physics engine that can check for collisions, depth of cut, speeds and feeds, material removal rate, tool geometry, etc and the results from that fed back into the AI to make another pass. Even then I would not trust the toolpaths to be safe or wasteful of material.
Yeah, it’s sobering how much data that involved — somewhere around here there’s a calculation I made about about how much data might be involved in just keeping track of cut/uncut material — current computers probably make this a bit more tractable.
There is one tool which @fenrus worked up based on my mis-apprehension of G-Wizard which I’ve been meaning to look into more deeply — it attempts to parse a G-code file and adjust feeds and speeds in it so as to adjust for various cases and arrive at a better cut.
That said, the last project where I pushed a CAM tool hard resulted in a >100MB toolpath file, and took some 20 minutes to calculate on an i7…
That said, I’ve been stuck on re-writing my current project — probably should make a post about that, since it sometimes organizes my thoughts…