They mention using a IMU dataset that is collected using an APDM Opal.
https://www.apdm.com/wp-content/uploads/2015/05/Opal-Publica...
This publication mentions a paper on p. 5839 (p 13 of the pdf) where a single sensor on the waist (as used in the Google research) would lead to an f1 score of 0.77 if I did my math correctly. In other words, pretty close to a >1 shot plot analysis of gpt4o and gemini pro1.5.
I would also be interested how the llm's would hold up to the free-fall interrupt that's built in to some consumer grade IMU's (BMA253 for instance), anyone here with experience in this usecase?
Yeah these ‘predictors’ only make sense if they can give a countdown at a constant rate.
The idea is nice but often they countdown at say 1dot/sonly to have the last 5 dots disappear in the last second so they miss their purpose.
on the other hand, a consequence on predictable ones is that people will start cycling on the last 2 dots or so instead of waiting for the green light.
I had a similar issue with geese on a patch of grass in front of the house where the kids played. Flocks of geese in the 10s would ruin it in half a day.
I found a 532nm green laser to be extremely effective. Let the dot approach them and shine near their feet for best effect. (Of course avoid the eyes and keep the laser power down so the reflection doesn’t harm them either!). It took some time to get into the flocks psyche. You had to get the entire group distress level to a certain level before they would take off and leave.
The next step I got to was building a turret for the pointer. Goose detection image recognition was underway to make the system autonomous but I haven’t been able to finish it and have moved in the mean time so it’s not likely I’ll finish it.
If anyone else wants to have a go at it let me know.
On the finishing:
“Of course because I 3D Printed it with a PLA filamanet, it's not as shiny and glossy compared to actuall electronic devices. People use acetone and various solutions to make it shiny.”
The acetone (vapour) approach only works on ABS and is pretty nasty. For PLA your best bet is sanding. You could add some putty to make it a bit easier and get better results. You will then also need to paint it. All in all if the print quality is good enough best stick with that.
XTC 3D is expensive for manufacturing but is really good for these hobbyist projects. It’s really too pricey to do actual production runs with but the results are great if you’re just making something for yourself at home.
If you’re using a color that already has a decent amount of yellow in it or don’t mind the yellowing polyurethane and spar-urethane are also decent finishing options. They won’t work on colors like white though, obviously. Even grey is probably going to see a little too much yellowing from those.
With overhangs/bridges, you eventually get into the habit of just trying to design a part so you can avoid that. Sometimes, of course, it is unavoidable. But it is quite strange and nice how your brain starts to just morph use cases into a shape and orientation that makes it easily printable after you've been thinking in the problem space for awhile. Eventually you just end up baking up creative printing solutions that try to avoid the need for supports entirely where possible because introducing supports introduces so many downsides from not only the material cost, but also the aesthetics, time, and frustration perspective as well.
XTC is a good solution for some use cases, and quite bad for others, as outlined above. For the use case in the OP, probably it would be quite nice. That use case is one that is hobbyist level of output in terms of number of units generated but also one where aesthetics are quite important if not fundamental to the overall design. For this kind of use case XTC is a good fit.
The one other thing to consider, if you design to use stuff like XTC and even the urethanes I mention above, is that obviously it takes up some amount of space. So you have to factor that into the design. Probably what I would do is take my boundaries from the original design and increase the tolerance by maybe 1mm or so for the extra space the finishing agent will take up. Because I highly suspect that if you applied XTC to the dock in the OP that the phone would no longer fit after that due to the tight tolerances in the design without that adjustment.
I have a Bambu P1S, and I'm fairly sure the prints would look worse after sanding than before. I just use them all as they are, straight out of the printer, and they always look amazing.
The primary thing people complain about is the aesthetics of the FDM print lines. They give the appearance of a hobbyist look. Sanding and finishing is one way to alleviate that. I bet PLA doesn’t sand amazing since it’s mostly just sugar but PETG sands reasonably fine. I have sanded PETG and finished with spar urethane/polyurethane for commercial products to give them a nice professional glossy sheen with no print lines. I mentioned in a sibling comment XTC-3D as another option which is probably the best hobbyist product for this, but it’s kind of a pain to work with and it’s a little expensive. So not great for commercial use. For home use though it’s totally fine and indeed usually gives better results than the urethanes mentioned above due to lack of yellowing and being thicker and better at filling in the lines and giving a smoother appearance. People could honestly probably just use that product without sanding and be mostly happy with it because it does a good job of hiding the FDM lines even without sanding
We run one of the largest print farms in North America. Reflecting parent's sentiment, BambuLabs' P1/X1 series are capable of substantially finer layer height and consistency (0.08mm), with SLA-quality results. We dumped all of our FormLabs because of it.
We don't sand any PLA, anymore. If we need to control the surface texture, it's largely a function of the qualities of the filament and print speed.
Great to hear. I have a resin printer at home and almost never use it because it’s such a messy, sketchy (in terms of chemicals) pain to deal with. The fact that FDM is good enough now for a lot of aesthetics use cases is great. I’ll probably still have to reach for the resin printer for high pressure use cases (like injection molds) but the fact that FDM can do more than before and make commercial grade parts is really good news
The physical characteristics of the print surface also change at lower layer height with high infill. I wonder if that might be good for your use case.
I'm mentioning it because we were surprised how easily our molds separated when we moved to Bambu and didn't understand why. It turned out that the combination of the layer height and high print speed that gave the prints their smooth, matte finish with standard PLA also made them functionally non-stick.
We also discovered they're also at least tough enough for a car to drive over, which may have just been because of the thickness. I also sometimes print nylon and polycarbonate automotive parts for a mechanic which seem to have excellent rigidity and durability.
Indeed I suspect FDM is good enough for a significant hunk of use cases as you’re outlining.
As FDM printing evolves the number of use cases it cannot solve seems to be shrinking. Probably we need to get into physics requirements to cover the requirements that FDM does not solve these days. Which is great, it makes general plastics manufacturing super accessible to the masses when we can arrive at this. I wish there was a general FAQ thing available people would be able to use to determine what manufacturing process would be necessary at the product planning phase. Because I feel like if people knew that they could just 3D print plastics at scale that a lot more people would be willing to innovate in the hardware space, which is a space well known as being difficult to enter and fraught with all sorts of monetary and regulatory land mines.
In my experience matte filaments hide the layer lines much better. Between that and the overall quality of recent printers like the Bambu or the Prusa Mk4, I don’t find the need to do any post processing.
I do want to challenge you on this. I don’t have evidence matte filaments are better. I would love to see an example of these filaments outperforming. I believe it is true, but would love to see a demonstration!
It isn’t really that they’re better as filaments, it’s just that the matte nature makes them less reflective and therefore you don’t see the layer lines as much.
Performance wise, most matte filaments are more brittle and have worse layer adhesion due to the matting pigment, but do a good job at hiding lines as the reduction in specular highlights reduce the visibility of them. Another good trick is fuzzy skin setting with both length and depth settings at 0.4x-0.8x layer height.
For PPE: an effective respirator is not that expensive, should be wearing them and safety glasses for sanding as well. Only extra PPE you really need for acetone smoothing is some nitrile / latex gloves, and those are fairly standard in shop / art environments anywho.
Also, MEK, Methyl Ethyl Ketone, apparently smooths PLA out quite well too, but if you can print in ABS I would already be doing that for finished products anywho.
Was looking for that too. The only way I found out was installing it, create a call graph, and the first click took me to a payment page. $19,95/month iirc
I too would prefer to have this information upfront.
Also, not being able to click past the first layer in the free version is a bit too limited in my opinion.
Sounds promising, but having to fill out my payment details for a trial version is a a hard no for me.
> The only way I found out was installing it, create a call graph, and the first click took me to a payment page
Yes, we need to be much more clear. It's not intentionally hard to find, we just been distracted with bugs and haven't had time to update the website. The subscription didn't exist until a couple of days ago. The layered architecture is a feature that we made public specifically for the HN community
> Having to fill out my payment details for a trial version is a a hard no for me
We'd love to embed a demo repo for people to play with, or let users generate nested architecture for one repo before having to subscribe
From the article: “ NextNav asks that the Commission reconfigure the Lower 900 MHz Band by creating a 5-megahertz uplink in the 902-907 MHz band paired with a 10-megahertz downlink in the 918-928 MHz band, shifting all the remaining non-M-LMS licensees to the 907-918 MHz portion of the band. Petition at 28–30.”
So at least all LoRa based applications seem unaffected by this. Though I don’t know how much traffic in the proposed NetNav bands will migrate to 915 MHz bands, crowding the spectrum.
"The USA LoRaWAN frequency spectrum has 64 uplink channels available (125 kHz each) (channels 0-63) starting at 902.3 MHz which increment every 200 kHz up to 914.9 MHz. There are 8 additional uplink channels (500 kHz each) (channels 64-71) from 903 MHz which increment every 1.6 MHz up to 914.2 MHz."
( https://www.baranidesign.com/faq-articles/2019/4/23/lorawan-... )
I would also be interested how the llm's would hold up to the free-fall interrupt that's built in to some consumer grade IMU's (BMA253 for instance), anyone here with experience in this usecase?