Hacker News Reader

Comment by csours | Hacker News Reader

Billion dollar question: do you get more bang for the buck (return on investment, ROI) out of improving robot control schemes, or out of designing the product with automated assembly in mind?

Bonus: the ROI changes as you invest in either bucket.

View on HN

That's a very interesting question.

Apple was once into design for assembly. The Macintosh IIci was Apple's peak at design for assembly. It was designed for vertical assembly. Everything clicks into place with a straight-down insertion move. No wiring harnesses. The power supply plugs into the motherboard. An automated plant in Fremont CA did the assembly.

Then Apple gave up on design for assembly and went to offshoring and cheap labor.

Motorola flip phones were designed for automated assembly. All parts were on boards, and the boards were stacked and compressed into a solid block, with bumps on the boards making connections to the next layer. A tough, reliable phone resulted.

Then Motorola gave up and went to offshoring and cheap labor.

Sony pioneered this approach. The Sony Walkman,, the original tape unit with motors and contra-rotating flywheels, was built for vertical assembly and assembled by a simple Cartesian robot.

Then came the iPod.

Do you, or anyone else here, know how much money Apple saved by offshoring?

I sometimes get it. Then when I hear robots were used, I wonder if it's really necassary to always go to the cheapest labor route.

For years, I held it against Apple for moving manufacturing, but gave up when everyone followed.

Manpower is a lot easier (and cheaper) to reconfigure between different products than robotic production lines.

> Do you, or anyone else here, know how much money Apple saved by offshoring?

There's another axis here, which is how our desire for a product overlaps with DFM. It could be the case that offshoring to cheap labor actually increased the manufacturing costs 2x, but enabled a product that would sell 10x better than its DFM counterpart.

(I have no data to say that is the case, only the intution that these things are complicated systems which rarely come down to single-issue decisions.)

For those unfamiliar, DFM = design for manufacturing. I had to look it up.

Apple case is not exactly abandoning design for assembly. The advancement of electronic and metal machining allows smaller and more integrated parts, which allows cheap labor to beat the machine. If the electronics and metal machining did not advance, I am guessing the resultant production cost would not be this low.

Machining is more or less in the same spot it's been since late 1970s.

I don’t think is true at all. CAM is dramatically more advanced than it was in the 70s - easier to use, and better algorithms mean much faster pathing. Costs are way down. Tooling is cheaper and more reliable.

Of course there's been quantitative improvements, but fundamentally everything that can be designed and machined today could be designed and machined in late 1970s using very similar tools, processes and control systems.

Yes, the technology is the same. But how many machined products were available for the consumer to buy in 1970? I was excited about the Macbook Air not because it was thin, but because it was CNC made, just like the aerospace products I designed. Injection molding remains dominant, but over the last 15 years CNC has made a lot of progress.

I dunno, by the end of 1970s most high end camera lenses were CNC machined with bezels and scales CNC engraved. My 1980 Summicron 35 is certainly one, although the optical stack sits in an injection molded envelope.

When it comes to MBA enclosure it's likely the infill or separation problems with long thin walls that led them to abandon injection molding. Even then it's not done on a mill in one take. Things like speaker grilles are probably something like EDM or etching rather than 8000 operations with 0.2mm drill bit.

You could not have done this with 1970s tools and CAD.[1]

[1] https://www.youtube.com/watch?v=Bqv5SjC4s6w

This was absolutely machinable with late 1970s CNC mill. Probably pain in the ass to design with then existing CAD/CAM but ultimately this demo runs on same G-Code that was there in 1970s.

And as far as parlor ticks go this is not particularly impressive, compared to old ones like turning cube inside a cube on a lathe.

That’s a bit like saying “most large modern systems are written in C which was invented in 1972, therefore software engineering has not changed since the 70s/80s”. That machine is also probably not running a gcode flavour that would have made sense in the 70s anyways, with additions for active sensing, macros, etc.

Tools like UNISURF would probably not be able to handle the extreme level of detail on most of that model. The very long tool lengths you see in the video are much more complex than they look and require effort both from software and hardware to prevent chatter and breakage on titanium. The clearances in the video are also extremely tight in places, and while you could have guess-and-checked that in the 70s, it’s a very different workflow than the simulators that are basically standard use today.

CNC machining is a multi-billion dollar industry populated by smart people. While the fundamental technology of “spinning cutter driven by computer controlled motors” hasn’t changed in 50 years, the R&D departments aren’t asleep at the wheel.

Machining grade 5 titanium alloy (the most common one) is not especially challenging; it's easier than 316. Saying this as I did machine it personally on a manual lathe.

From the initial endmill size they use and the generous machine sizes (which can be extrapolated from chuck/table size) we can easily tell rigidity/chatter wasn't a particular challenge here. And the finishing passes are so delicate they don't even bother with coolant for that.

Yes the code to run all this would be a monumental undertaking in the 70s but it is done once and can be still entirely cost efficient as far as mass production is concerned. The thing new machines have is the running speed (for same precision work). These did improve a lot and make certain classes of products economically viable.

Yes, in the 1980s people were told they'd lose their manufacturing jobs to Japanese robots but the robots turned out to be Chinese workers.

Half of automating tomato harvesting was breeding a tomato that could survive the harvesters. (Granted they also taste relatively horrible, but now we all eat them because they are cheap and ubiquitous.)

That sounds fascinating. Do you happen to have some details, sources, videos, lectures about this?

Here's a short article:

https://www.plantsciences.ucdavis.edu/news/how-mechanical-to...

on edit: I also went ahead and made that article a seperate post here:

https://news.ycombinator.com/item?id=27937200

The improvement of robot control schemes reduces the constraints on designs that are designed for automated assembly. I suspect there’s a kind of slow moving coevoultion there where you have to go incrementally at both.

Probably both, to the extent that it's practical.

Indeed. Looking at their sample footage of assembling Ikea furniture reminds me of fixturing. Watch manufacturing footage and watch out for jigs and fixtures. They are EVERYWHERE.

Currently, you can either use fixtures and jigs and specialized machines and run fast, use humans and run medium speed, or use AI and generic robotics and run REALLY SLOWLY.

Where's the value prop?

I'd say invest in the former, as having better robot control schemes should allow you to more easily iterate on different design alternatives

Reply on news.ycombinator.com