HN Theater @HNTheaterMonth

⬐ Dig1t

I think illustrating how everything fits together with an animated 3D model is extremely underrated. I wish there were more videos explaining all kinds of concepts using this approach. So much information is conveyed so quickly with this spacial representation, though it's probably a lot of work to produce videos like this.

⬐ sfteus

While the animated breakdown itself is phenomenal and certainly makes the video, one of the other key aspects is the progressive explanation of _why_ this mechanism is designed the way it is. You can watch the video without sound, and probably without the text as well, and see exactly how the "spikey ball" was designed, how the driver gears were created, how to get two types of movement from the drivers from linear inputs, and how those movements translate to moving the ball joint.

It reminds me of the old Chevy videos, such as the one on differentials[1]. It was created in 1937, and through some live demonstrations and clever use of stop motion the film shows how to separate wheel movement, fix gear slippage, attach a drive shaft, then optimize for space. Different visual technology, but same type of presentation. There's similar videos for transmissions, suspensions, etc, all incredibly enlightening.

[1]: https://youtu.be/yYAw79386WI?t=202

⬐ quakeguy

You may find this channel interesting, all animations are selfmade by him: https://youtube.com/user/thang010146

⬐ adamrezich

totally, I played the video without sound so I don't even know if there was any verbal explanation but if there was it was unnecessary, the visuals conveyed everything perfectly.

⬐ trevcanhuman

I watched the video and there wasn’t any sound. Definitely a step by step graphical explanation helps a lot.

⬐ eco

I came across the YouTube channel of Jared Owens[1] recently which is basically just that.

1. https://www.youtube.com/channel/UCbsfyGlrjrKQC0gbzK0-EiA

⬐ Dig1t

Wow, this guy is amazing, thank you for sharing this!

⬐ colordrops

I was about to reply with this guy. There's no comparison, he's light years ahead. He can make any complex system comprehensible on first pass. Watch his video on bowling pin setters, it's amazing.

⬐ addaon

Apologies for the content-light comment, but this is awesome. Interesting mechanism, well illustrated, and taken through completion with integration into a module, not just a single mechanism.

⬐ tejtm

Very nice.

One perhaps counterintuitive thing about threads and gears is the optimal "tooth" size is a function of the material strength, not the geometry of the object the tooth is on.

Another is that when regular involute gears mesh, they press but do not rub, no sliding friction.

Here I am not seeing how to avoid sliding friction which is a small price to pay for the extra degrees of freedom but one to factor in.

⬐ codesnik

I also see some locking positions which probably would require active rotation of the supporting gears in their brackets.

⬐ rfrey

Slightly know-it-all, but in fact involute gears do rub against each other - they experience pure rolling motion where they contact at the pitch circle. That's the only point where the circumferential speed of the gears is the same.

The purpose of the involute shape is to maintain constant torque transmission during tooth contact.

⬐ jerome-jh

I believed like OP involute gears had true rolling motion with no slide however quoting Wikipedia (https://en.wikipedia.org/wiki/Involute_gear):

"Where the line of action crosses the line between the two centres, it is called the pitch point of the gears, where there is no sliding contact."

So I learnt something :)

I also believed torque varied slightly because the point of contact between teeth was not at a constant distance of the gears centers. Illustration: https://en.wikipedia.org/wiki/Involute_gear#/media/File:Invo...

Would you have sources for the affirmation "constant torque transmission during tooth contact"?

BR

⬐ rfrey

I overspoke when I said "constant" - like everything, involute shape is a compromise, in this case (I think) between constant torque and manufacturability.

I should have said "the shape is intended to minimize torque transmission ripple".

I'll pull out my old textbook and see if I can figure out where I picked that up from.

⬐ jerome-jh

OK thanks actually I do not need the math. Just wanted to know I my beliefs were wrong on that point too.

⬐ _Adam

This is super cool and the video explanation is very intuitive. Robotic manipulators seems like the obvious application; I wonder how the torque transmission compares to a more traditional arm design.

⬐ baybal2

PTZ cameras

⬐ convolvatron

does anyone understand why we have 4 drive motors for 3 degrees of freedom? it didn't seem that way from the presentation but maybe the motor axes aren't aligned with the drive axes?

⬐ zardo

It eliminates gimbal lock

⬐ Animats

No, it doesn't. Watch the video out to the end, and you'll see what happens as you go through a pole. This isn't a homogeneous system; there's a moment when the gear flips.

⬐ jbay808

That's an accident of the trajectory chosen for the animation. The system doesn't have a kinematic singularity from input to output, just from output to input. Instead it has a zero from input to output, which is why it needs an otherwise redundant motor.

To elaborate, there's one point where spinning an input motor doesn't rotate the output. When they reverse-generated the input trajectory from the output trajectory, that point has no unique solution; spinning the input motor has no effect on the output so the input can do whatever it wants. The animator let the input do a sudden backflip, but it didn't have to; that motion was not required and had no effect. It was probably just the output of a matlab script or something.

⬐ jbay808

Two of the motors have to be synchronized together, because there are certain angles where either one or the other lose any torque transmission.

⬐ chrisBob

This looks amazing, but it isn't 3 full degrees of freedom is it? I feel like there are some orientations that wouldn't be possible with this, but I am really not sure.

⬐ baybal2

You can make a spherical induction motor

⬐ chrisBob

The paper (Open Access!!!) https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=941... Says that I am wrong, and I am willing to trust their analysis.

⬐ R0b0t1

The retaining cusp does impose some restrictions.

⬐ rtkwe

Those aren't usually counted in enumerating the degrees of freedom of a robot arm unless it's a significant restriction.

⬐ aj7

Here:

https://www.researchgate.net/publication/351357682_ABENICS_A...

⬐ Animats

Oh, that's clever.

They have to coordinate four motors to get three degrees of freedom. Not clear what the invariant is, but it may be something like a normalized quaternion.

Mechanically, all the load is on maybe two tiny teeth at a time. This isn't going to be an industrial robot leg joint. Or, probably, even an arm joint. Too easy to strip the teeth off the sphere.

⬐ gugagore

I don't think the constraint is like normalization.

Consider a platform with https://en.wikipedia.org/wiki/Omni_wheel

There are 3 degrees of freedom for rigid bodies in the plane. If you have four wheels, then there is a constraint.

Associate with each wheel a unit vector along the direction it can impart force, perpendicular to the direction that it imparts no force. Now take a vector indicating the velocity you want to travel in (ignore rotation for simplicity).

To figure out how the velocity of each wheel, take the dot product of that wheel's unit vector with the target vector.

To see that normalization doesn't play in the constraint in the omniwheel case, note that any valid assignment of wheel velocities is still valid if you scale it up or down.

I think the case here is more complicated because it's not a euclidean space. There are poles. I believe underlyingly my analogy holds, though, if you think about manifolds and tangent spaces.

⬐ Animats

I don't think the constraint is like normalization.

Yes, this is different. I was thinking of a 3DOF trackball setup. There are 3DOF trackballs (you can both roll and twist) where a big trackball rests on three small ball-bearing balls. Two of those balls each have 2 axes of rotation sensing; one is just an idler. It's all friction-based; no gear teeth. So you have four incremental data signals coming out, which resolve to the three motions a sphere can do. That approach is homogeneous; no position of the trackball is special.

⬐ holoduke

Wonder whether the non driving gear needs to be aligned with the sphere. Or does the force push the driving gear into a gear alignment? Or is it done in software. I noticed some jerkiness in some movements. Seems that in some cases gear play is definitely there. Not good for precision. A very cool design though. Makes me want to 3d print it

⬐ dialogbox

This is very cool. However I'm not sure how much torque the ball can endure. All gears have to be really strong and very precise. Is it really practical?

⬐ amelius

> Is it really practical?

Not sure. One application would be the last stage of a robotic manipulator (at the end of an arm), but I suspect it's really too big for that.

⬐ bserge

Tbf, gearbox shafts and gears can handle a lot of torque and power for their size.

⬐ knodi123

It can handle a pre-determined amount. Just like all gearings. :-)

There's definitely a tradeoff here, but I imagine there are plenty of applications where it makes complete sense.

⬐ Aspos

I guess this would make a fast, precise, and optically centered pan-tilt mount for a camera.

⬐ warrenm

I think my brain just broke watching that

Very cool!

⬐ nope96

wouldn't this need some sort of lubricant, or it would wear down over time?

⬐ everyone

Awsum!

⬐ gfodor

This makes me wonder if ML could be used to explore the space of threadings to optimize torque or reduce risk of disengagement. Maybe even drop a motor.

⬐ antegamisou

Because ML (and NNs ofc) is definitely a one-size-fits-all solution to interdisciplinary problems..

⬐ gfodor

What happened to you that you decided replying to this with a emotional strawman was worth your time and energy?

It's a particularly dumb strawman too because we already know AI can generate solutions to mechanical engineering problems that humans normally would not. https://medium.com/intuitionmachine/the-alien-look-of-deep-l...

⬐ antegamisou

It's generally a bad idea to invest too much in ML methods for physical world problems, especially considering it a panacea when their mathematical foundations are still poorly understood. The cost may be only computational when it comes to areas like Image Processing/NLP, however it's nowhere near the same for things like AVs (safety), Engineering Design problems (materials) etc. And this is because real world imposes real hard constraints, to the point that it'd be unfair to expect similar success to CS-related disciplines here from ML methods.

This is no different for manufacturing problems. Excluding the absurd PoCs/artworks, most of the actual structures in the article you've linked are impossible to mass-manufacture without 3D printing, which is still limited to precisely printings parts with unsuitable materials for their target application.

Keep in my mind that I was mainly referring to applying emerging trendy methods for which mathematical guarantees have not yet been established. Genetic Algorithms, for example, have been able to come up with successful antenna design optimizations like the one in the article for almost three decades.

⬐ gfodor

“Panacea” <— your strawman

I noticed a lot of your replies are similar in talking down to people. Weird!

⬐ cyber_kinetist

I generally agree with your line of thought (CS is too accustomed to toying around with imaginary models/rules and not real-world physical ones), but I don't get the connection between mathematically sound models and manufacturing problems. Are even Genetic Algorithms (a bunch of metaheuristics) really mathematically that robust? I thought GAs were mainly discovered and developed purely empirically, and there's not that much theoretical guarantees behind it even as of today. The reason it's being used in the context of engineering is that it seems to solve non-convex or combinatorial problems really well compared to previous approaches, and that's it. And if GANs or Deep RL can explore solution spaces more effectively than GA algorithms, then why not?

Many engineering problems can be reduced to optimization problems, and if a certain toolset can be used to effectively solve these optimization problems, then it should be applied pragmatically. You could still question if ML brings anything new to the table ("isn't it just curve fitting in higher dimensions?"), but I think there's some value in using data-driven ML toolsets judiciously. For example, a lot of engineering tasks are repetitive but vary in very slight deviations (like screwing things to holes), and for these cases statistical models might help in bringing efficiency and safety. Of course the system still cannot handle real catastrophic deviations from the norm (ex. small earthquake happens and the robot is shaking) - but at that point human intervention should definitely happen. A truly intelligent AI would handle even the exceptional cases with some degree of pragmatic decision-making, but that's not the goal we're chasing for here.

I still agree with the sentiment that there needs more work to make ML methods safe though. This is in two grounds: one is to make it easier for the programmers to add domain-specific hard constraints to the model they're learning (enforce the output of the model always adhere to certain rules), and the other is to make the ML system report to the human operators for manual intervention when any abnormalities to the input and the output are detected. It's a bit disappointing that much of "safe" ML has focused on ethical issues mainly arisen from Silicon Valley (like gender/race issues in models: duh, your model will probably have the same bias as your data! The conception of using ML for solving social problems is fucked up in the first place...). The real focus should be the ability to steer, control, and constrain ML models as human operators would like, and I think the neglecting of this is the reason why ML has largely failed to be incorporated in industrial domains outside of SV as of today.

⬐ antegamisou

Thanks for this thoughtful comment.

It would be dumb to say that it's not worth it to pursue ML research and apply it to engineering problems to improve well-established solutions.

What bothers me the most however and is the reason I was tongue-in-chick in my first comment was the insane overrepresentation machine learning receives in virtually every interaction related to tech, considering its success to non-physical problems. I get that hypes come and go and will continue so, but it's another thing having a trend spreading out in almost every scientific discipline advertising it as a discovery by big brain computer people that have come to salvage their poor inferiors (that its ardent proponents have zero experience in engineering design is another story). You may think this is hyperbole but it's really a common sentiment among ML skeptics. And it's especially frustrating when it's presented as a subtitute, and not as a supplement, to well established mathematical ways, e.g. machine learning vs control theory in AVs as you've said.

On the other hand, although SV startup culture is definitely to blame for all this to an extent, I can definitely understand some subfields being way too conservative to trendier topics just for the sake of not blending in with the hype. There have been some decent attempts lately to bridge the numerous chasms each discipline has and get the best out of each world. Hopefully something useful comes out of it.

⬐ gfodor

Statistical learning of functions is as general purpose as structured programming. Your “voice of reason” take will be proven as enlightened as those who swore up and down that structured programming was a fad with narrow applications. It’s a tool like anything else but given the widespread accessibility, utility, and commoditization you’re behind the curve arguing with strawman proposing it is a “panacea” and not just a core part of your engineering toolkit.

⬐ theelous3

I got in to machining at the start of the pandemic, I suppose just short of two years ago. Absolutely brilliant hobby.

I remember a comment here a while ago about a lad who was interested broadly in systems and diagnostics, and had initially aimed to be a doctor. They then discovered they wanted to work on systems designed by logical first principles, and pivoted to comp sci and programming - only to find they'd discovered a whole new kind of almost random organic system.

I think machining is about as close as it gets, in terms of the physical. The depth to the subject is off th charts. It all logically follows from first principles ;everything is rubber ;D

It has an incredibly satisfying balance between the theoretical and the applied.

Physical mechanics is a truly beautiful thing. Doing it yourself is equally fascinating and fun.

Can't recommend it enough.

⬐ robomartin

> everything is rubber

I've been saying this for years. I go one step further, I call it "resonant rubber" --a mythical material that actively conspires to evade everything you are trying to do to it and has a mind of its own.

⬐ theelous3

Ha! Yes, even more accurate. I just motorized my quill feed, and today's plan is to find the speed it resonates at, and program it to skip that speed band.

⬐ robomartin

While we own a full machine shop at our company I still have a very, well, unique, garage. At the moment I have my old manual Bridgeport knee mill and a Haas VF2 right next to it (along with a prototype-scale SMT line and other goodies).

I've been going back and forth with the idea of turning the Bridgeport into a CNC/manual hybrid. I like the idea very much, but I also love full-manual machining. I've used such a hybrid in the past, with both a motorized knee and quill. You could still disengage the quill and simply grab handles and machine manually.

The control they used was the Acu-Rite Milpwr G2:

https://acu-rite.com/controls-readouts/millpwr-g2/

It's really nice, but fairly expensive (I think about $25K by the time you are done).

One of the arguments I read against converting a manual mill this way is that the ball screws create a situation where manual machining can be more difficult due to the low friction and back-drivability. A machine builder told me it would be a better idea to trade it for a Bridgeport Boss, which is actually designed as a CNC machine.

Having the VF2 next to it I don't generally feel the need to make lots of parts on the knee mill. Like it or not, it makes a mess. That's why I keep thinking that the right level of automation might be to only motorize the x and y axis for quick and accurate hole location and simple machining operations and leave the quill and knee fully manual.

⬐ diego898

Awesome! Can you recommend some intro resources to help someone get started? What worked for you? What didn’t?

⬐ rfrey

The youtube channel "blondiehacks" is excellent for the machining-curious.

⬐ aj7

Yes that’s where to start.

⬐ emmelaich

https://www.youtube.com/c/Blondihacks

⬐ theelous3

For learning to love it in an approachable way, this old tony https://youtube.com/c/ThisOldTony

For more practical stuff, joe pie https://youtube.com/channel/UCpp6lgdc_XO_FZYJppaFa5w

For the very precise, robrenz. https://youtube.com/c/ROBRENZ

But my favorite - stefan gottswinter https://youtube.com/c/StefanGotteswinter

Take a shop tour with stefan and nycnc: https://youtu.be/H-Sf7Nvkwzg

⬐ gaze

Robrenz is the favorite machinist of all the YouTube machinists. He might be one of the best in the US.

⬐ mmaunder

Agreed. I got into cnc. I use a Shapeoko 3 and 4. Both are xxl. Started on wood and then transitioned to non-ferrous metals. First aluminum and now copper alloys like bronze. Metal is a big step because you really learn about feeds and speeds. I use Fusion 360 which costs but it’s the best IMO. Amazing intersection of engineering and art.

⬐ rfrey

The other appealing (to me) thing about machining is that one simultaneously: (1) is forced to realize that perfection is completely illusory: everything is made of rubber, there's no such thing as an exact dimension (2) gets as close as any human endeavor to actual perfection. An amateur can, with care, skill and some money, work to microns.

⬐ theelous3

Aye. I have a pretty cheapo mill, 2k for the mill and 6k all in (most expensive sector is metrology)*

Can easily work to under 0.01mm which is absurdly small when you're faced with it.

Then you see the guys working under 0.0002/1 and... it's just incredible.

> gets as close as any human endeavor to actual perfection

Truly. The only thing that matches it in terms of the physical is Xnm chip production and atomic level physics research. But those are far less hands on, and require tens of millions in equipment (plus the parts areade by machinists anyway).

* can provide running costs breakdown if interested

⬐ rfrey

> (most expensive sector is metrology)

Absolutely. But ohhhh... those instruments. I saw a friend with an electronic gauge put it on one end of a 1' bar of steel. He put the gauge on one end and zeroed it. Then he breathed on the other end. We watched the needle cliiiiiiiimmmmbbb.... then fall back to zero over 3 minutes or so.

⬐ tomcam

Holy crap. Is this because the steel expanded an infinitesimal but somehow still measurable amount?

⬐ theelous3

Yes. There's a joke in machine shops working to insane tolerances.

"just breathe on it till it measures in spec"

There's a great video by oxtoolco on youtube where he's seeing which is "thicker". A sharpie mark or some machinists blue. Measuring in the millionths of an inch.

⬐ rfrey

Spoiler: a sharpie mark is about one ten-thousandth of an inch thick (depending on color, because size of the pigments!)

An interesting use of this bit of trivia: when machinists are trying to grind a block square, they may measure a slight deviation from 90 degrees - i.e. they see that each of the two opposite faces are parallel, but the angle they make is out of square. They have a parallelogram.

If you draw it, you can see that if they put it back on the grinder, and shim one edge of the face contacting the table, when they grind the opposite-to-the-table face they can bring it closer to square because they'll be grinding a wedge off the top face.

They use a quick swipe of a sharpie marker as a shim.

⬐ rfrey

Yes exactly. I was blown away (ahem) and instantly regretted my life choices, which until that point had not involved microns in any way shape or form.

The instrument readout looks like this: https://www.perfectionmachinery.com/assets/item/hires/40961_... except his had adjustable ranges. Every tick on that needle is one millionth of an inch - by comparison a red blood cell is about 40 millionths and the smallest bacteria is about 5 millionths long. This instrument has a read head connected with a cable, that has a finger on it that rests on the part under test. Under normal use you'd slide the part around under the finger to look for variations in thickness - obviously that requires and extremely flat surface to slide around on.

⬐ falcolas

> An amateur can, with care, skill and some money, work to microns.

I love how you can start with really rough tools (like wood and stone tools), and iteratively create more and more accurate tools, to the point of rivaling commercial tools.

⬐ kaba0

Could you point me to some resource on how would such a process look like?

⬐ rfrey

Start with researching the "three plate method" to generate a flat surface. A flat reference plane is the first thing you need to make precision machines. (The second is an accurate way to divide a circle.)

But the three plate method is one of those foundational, from-first-principles realizations that changes everything. It basically means you can start with three rocks and manufacture your own reference plates that are accurate in flatness to any arbitrary tolerance you have the patience to achieve.

With a truly flat plate in hand, you can generate a reference square - again, to any tolerance you like.

If reading about that stuff (it's everywhere on the web if you know the right codewords "three plate method"), your next step is to find a copy (electronic, probably, it's long out of print) of Moore's "Foundations of Mechanical Accuracy". I keep a copy beside my bed so on nights when I can't sleep, I can caress the covers and ease my soul.

⬐ kaba0

Thank you very much!

⬐ edoceo

I mean, that's how we got here. In the grand scheme the time from using rocks to make fire (flint) to putting electricity into thinking rocks (silicon) to throwing refined carbon into space was very short. Amateurs stacked on amateurs from only 1600 mothers ago!

⬐ robbedpeter

It's only been 340-ish years since we started on the electricity tech tree in any rigorous sense. That's 16 or 17 moms ago, at most. If you had a durable, stable culture 200,000 years ago and dropped a series of carefully selected experiments in their laps, they'd have been able to create with electricity with as much brain power as we have today.

Still, we had thousands of years of cultural experiments to survive to eventually arrive at a confluence of social structures that allowed for us to explore science and rational natural philosophies.

From the Greek mathematics and Sumerian city states to Roman empire building and onward, we carry pieces of those things that perform best at giving us life less full of struggle and suffering.

A smartphone, seen from Kurzweil's view of human progress as a function of structured information, you get a sense of being present for something vast and terrifying and beautiful. Stacked amateurs all the way back to the brave little rodents thriving after the dinosaurs died.

⬐ vimax

That reminded me of the poem The Calf-Path

https://poets.org/poem/calf-path

⬐ robbedpeter

Thank you!

>>They follow in the beaten track, And out and in, and forth and back, And still their devious course pursue, To keep the path that others do.

⬐ mikewarot

As someone who made gears for a living, it's rare for me to see something new... this looks great! If it works long term, I see a lot of interesting new robots coming down the pike.

⬐ rektide

4 upvotes yesterday for the paper[1]. I am fascinated by this thing. Awesome.

[1] https://news.ycombinator.com/item?id=27554760