Firstly, at the $219 price point you can have my money already.
Beyond that, things that appeal to me are basically anything which increase the likelihood I can accomplish high dexterous fine motor control skills, for things like tinkering and DIY assembly. I think that would include extra wrist DOF and a longer-reach variant.
Integrated cameras are an interesting idea, but I'd like to be able to swap them out for my own.
My dream is to have some sort of multi-arm table at home. I imagine holding a circuit board, small component, soldering iron, and wire with four robotic arms I control with shaky hands from my laptop. :D
So true. Every time I solder surface mount components, I always wish I could have a steady hand. Sadly, this arm doesn't have that kind of accuracy. The output shaft of the servos we use has about 1 degree of wiggle room and the mechanical structure adds more.
To get better accuracy, if sticking with this kind of RC servo, it's basically required to have two servos per joint to preload each other to kill that wiggle room. It's something I've been calculating, but I just can't figure out a way to offer it at a good price.
Interestingly, for arms that are popular in academia, even when the price goes to $10k (like ARX or Trossen), the wiggle room is still there (better, but still there).
I was recently trying to get better angular accuracy with servos and minimize backlash. One option that kind of worked was to have a pulley on the servo shaft which wound a string attached to a spring to add mechanical bias.
But I ended up giving up and going with 400 step stepper motors instead. They're larger, draw more current, and the drive circuitry is more complicated (it can't get simpler than a PWM servo after all). But they're accurate and significantly quieter.
Oh, I think I actually saw this! But it's a very manual process with a lot of small parts to assemble, and I want to eventually design a product I can manufacture myself so I didn't pursue it.
What you need is to scale down the tolerances. To remove the wiggle room.
One of the solutions that does not add a bias that I remember is two identical flat gears on the same axis with a spring that tries to rotate them one relative to another. This removes the wiggle room between this composite gear and the next, regular gear. The motor may have wiggle room, but the gears (which carry angle sensors, don't they?) move without wiggling, and react immediately as you reverse the direction. The load is limited though: the beating surface is twice as small, and the friction is higher.
One could put those in series too and get even less range of motion in exchange for less wiggling.
One could also duplicate the contraption on both sides. Then could replace the arm with cables (under tension) and control motion further down the arm.
Furthermore it seems you could remove the motors from the moving parts?
You could also control the tension of the wires and have accuracy on demand without the wear. I have this mental image of a darts player moving his arm back and forwards to smooth out the tension.
I sometimes wonder if it makes more sense to just use those yellow gearboxes, everyone seem to start with SG90s only to reimplement most of the servo part.
have to share the noob thought because it is funny: You could attach an unbalanced wheel to a motor and induce a vibration to maximize wiggling and frequency across the available w-room.
Backlash is a real problem. I’m working with the supplier to prototype a small batch with 20 to 30% less backlash, though this may come at a slightly higher price and potentially a shorter lifespan.
If cost isn’t a concern, harmonic drives combined with brushless servos are excellent. I have a few harmonic drive units, and they’re truly amazing.
There are cheaper options for soldering SMD-parts on prototype boards. Developing and teaching robot arms to do it would give a good demonstrator but economically it’s a disaster. And mass production is already highly automated.
even something twice the price ($438) would still be a great deal. Mind telling us something about your pricing strategy trade-off consideration matrix?
Design for manufacturing is one thing. I did it a lot when building micro gas turbines in college. Sometimes changing the design or manufacturing process will make it 10x faster to make one while not compromising the performance.
The second thing is low margin. When people are pricing hardware, they usually plan a 50% to 100% margin to offset various costs that happen in the real world. From what I've heard, in extreme cases, some products cost around $100 while they are being sold at close to $1000. I believe in the Prusa printer approach: you design a good product and price it a little bit above cost. So the company grows with the community.
Deep down, there are so many times that I wish I could afford a fancy tool like a Milwaukee drill or a Mitutoyo caliper. And in extreme cases, I really wish I could have a HAAS UMC-400 or even a KERN Micro HD+. Now that I can set the price, I really wish I could make someone get what they want without breaking their bank.
That’s not at all reassuring to someone who wants the benefit of the product, not the obligation to become an expert on the product.
I’ll buy one of these for sure, but I would cheerfully spend 3x the price if it meant being sure of support and repair and software updates for a few years.
I have bins full of exciting devices that no longer integrate, many of which have open source communities, which I just don’t have the time to deal with.
This is based on the open-source HuggingFace/LeRobot SO-101, so it'll probably be possible to fall back to that. Or you can just get an SO-101. But I think this is attempting to draft on that ecosystem, and this is cheaper than Alibaba parts to build an SO-101.
that's more scary, not less! Sure, if the worst happens, since it's open source, we'll have the source, but being open source means they've got to figure out a more complicated business model than make thing, sell thing, profit. there are some success stories but also a lot of failures.
I guess I should clarify that it's open source, and seems to be essentially forked from HuggingFace's SO-101, so it's probably not too hard to fall back to that stuff (with fewer features).
It’s worth reading the history behind Raspberry Pi. It depends on the team. With the right product market for, this strategy has and will work. In my view, the Vassar team got the pricing strategy right to do something even bigger than Raspberry Pi if they can scale at this price point for the low end model.
The advantage of servos is that they can provide constant torque preloading, allowing the preload to be kept low (otherwise, the servo will overheat) but still sufficient.
A spring might also be an option if designed properly. I’ll probably give it a try in July.
You can use a spring with long travel and low K (like a very soft torsional spring, preloaded several turns, or a spring connected by a wire to a drum on the joint) to reduce the change in torque across the range, if that helps?
Yeah… the accuracy isn’t good enough. A gantry system is probably needed.
There was one robot startup (Haddington Dynamics) figured it out how to do it at a higher price. Sadly they've been acquired and shifted directions I think.
I had that same thought. I travel a bunch for work and would love to be able to wire up a RPi/Arduino remotely to a sensor or other device and run a test or two.
Ooooo I will definitely check it out! It's strangely hard to find any comparisons in youtube videos -- it seems TDA isn't actually a dimensional reduction algorithm, but something closely relayed, maybe?
No other book captured the feelings of being 20-something and flirting like reading this. Reading it felt like being right back there again, with all the excitement and anxiety. Highly recommended to anyone.
Unsure how it connects to the notion of a brain filling in the blanks. I thought it was quite "filled in", but maybe my brain did it, and therefore I'm making your point for you :)
I look forward to reading this in closer detail, but it looks like they solve an inverse problem to recover a ground truth set of voxels (from a large set of 2d images with known camera parameters), which is underconstrained. Neat to me that it works w/o using dense optical flow to recover the structure -- I wouldn't have thought that would converge.
Love this a whole heck of a lot more than NeRF, or any other "lol lets just throw a huge network at it" approach.
Thanks for what you're doing. Of all the various companies and orgs posting chatter about deep learning, I've come to really appreciate your efforts (and Anthropic), because you're USING MATH. :)
I have some understanding of applied math, continuous and discrete, and while I don't keep up to date with developments in deep learning/AI in general, I always look forward to unsloth posts because they tend to center on achieving a desirable result thanks to proper application of good old fashioned "understanding the damn math and once you do, then doing the obvious". :)
Reminds me of learning about how to optimize twiddle factors in wring a performant FFT (more than divide-and-conquer, one also uses trig identities and some algebra to reduce the number of multiplies), or of learning of elliptic minimal Q-factors (EMQF) filters -- clever IIR filters that give a sharp frequency response using less than 50% (or is it 25% or more?) of the computation required traditionally by optimizing for *coefficients with lots of zeros in the base 2 representation*. And computers, it turns out, can multiply numbers by zero really fast. ;-)
The throughline to me is that if you pause and think deeply about "wait, what are we really doing here?" and look at the whole math stack, and think about what computers are good at, sometimes you can achieve great results.
I always keep maths at the center of everything we do :) It's literally humanity's ultimate superpower if we can express everything in mathematical terms!
This is awkward—I use em-dash all the time on HN! I'm not an LLM (as far as I know); I just like to write neatly when I'm able to, and it's very low friction when you're familiar with your keyboard compose sequences[0]. It's a trivial four keypresses,
AltR(hold) - - -
(The discoverability of these functions is way too low, on GNOME/Linux; I really dislike the direction of modern UX, with its fake simplicity, and infantalization of users. Way more people would be using —'s and friends if they were easily discoverable and prominently hinted in their UX. "It's documented in x.org man pages" is an unacceptable state of affairs for a core GUI workflow).
never knew about the em dash thing, I was just using an AI writing assistant to help fix my shitty grammar and formatting. I think in future ill stick with bad formatting
Wow! This is really cool! Really really cool! I imagine some sort of use where it's even more collaborative and not just "unadorned turn-by-turn".
For example, maybe I'm taking notes involving words, simple math, and a diagram. Underline a key phrase and "the device" expands on the phrase in the margin. Maybe the device is diagramming, and I interrupt and correct it, crossing out some parts, and it understands and alters.
Sorry, I know this is vague, I don't know precisely what I mean, but I do think that the combination of text (via some sort of handwriting recognition), stroke gestures, and a small iconography language with things enabled by LLMs probably opens up all sorts of new user interaction paradigms that I (and others) might be too set in our ways to think of immediately.
I think there's a "mother of all demos" moment potentially coming soon with stuff like this, but I am NOT a UX designer and can't quite imagine it clearly enough. Maybe you can.
Yes! I have flashbacks to productive times standing in front of a whiteboard, alone or with others, doodling out thoughts and annotating them. When working with others I can usually talk to them, so we are also discussing as we are drawing and annotating. But also I've handed diagrams / equations to someone and then later they hand me back an annotated version -- that's interesting too.
