>>44379 CHONK!

>>44380 >CHONK If being a little thicc is what it takes to keep SPUD from exploding on impact, bring on the thighs!

>>44381 /thread

SPUD leg is almost complete: it even stands by itself. I just need to reprint a few parts as the linkages need more clearance to fully extend. I'll also probably cut down the width of the metal linkages, since they're made out of 3/16" steel. That big piece of angle iron was part of a bundle of materials donated to me by an irl fan a few years ago. Also due to tolerances this will probably end up being the left leg instead of the right. At least I printed both feet so swapping should be easy :D Then I suppose I should design some feets to hold the batteries. That's gonna add a few more inches to her 5' 4" height.

>>44385 POTD Really proud of you, Mechnomancer! Looking forward to your further refinements on these and the operational prototyping. Have you figured out some kind of gantry + harness type thing to facilitate your walking research yet? Its very cool that the individual legs can stand up alone. Keep moving forward.

>>44379 >>44385 Nice, it's looking really professional!

@Mechnomancer This video mentions a smol SPI-to-CAN converter board; both topics we've discussed before. Thought you might be intredasted. Cheers, Anon. :) https://inv.nadeko.net/watch?v=6Lgqm0KW-0g

>>44401 >Open source robit with manual >cloud-based control program >shite documentation with no step-by-step instructions just explanations of random-ass concepts >pretentious robit name of "asimov" >like a 3 foot midget kek I consulted t̶h̶e̶ ̶o̶r̶a̶c̶l̶e̶ chatgpt and got recommended MuJoCo which seems to be an actual robit walking training program (so no hallucinations there). I'll be looking into it to train SPUD to walk. There hasn't been much updates to the leg other than getting the knee to actually bend. So I'm starting to print the entire next leg.

>>44410 Lol, true. :D >MuJoCo >I'll be looking into it to train SPUD to walk. Good luck! Eagerly looking forward to a positive report, Mechnomancer. >other than getting the knee to actually bend. I don't take that lightly, and in fact I'm excited by the progress. Cheers Anon.

You should really update your website, it still has "Lorem ipsum dolor"

>>44567 >update website yeah, I know After printing out the entire 2nd leg I realized I need to re-design the knee servo so the 550 motor on it faces out, not in. D'oh!

>>44571 Nice work! It looks just like a "professional" humanoid robot! *Professional being in quotes because I do not think that institutions are inherently superior

>>44571 POTD Looking forward to this Anon. Cheers.

Mechnomancer, I made a thread about IRL mechs on /mecha/. The thread is like /robowaifu/, but for mecha https://trashchan.xyz/mecha/thread/1075.html

>>44583 Yeah I saw the thread. I could waffle for hours about IRL mech design but I don't have much free time atm. :)

>>44583 Are you going to add that thread to our new "listings" thread? (cf. >>44504 )

>>44587 Sure. I initially made it for just Trashchan /robowaifu/, but we can add our robowaifu adjacent and robowaifu embassy threads

Its been a while, been doing other stuff that only slightly has to do with robits. Tried a diy silcone face. TLDR: functionally successful but ugly AF. More detail: spread bathroom caulking on a positive mold, using toilet paper as a separation medium. It turned out better than expected: I was able to (gently!) separate the silicone from the mold without any damage (letting the piece cure for 48 hours) and the material is functionally usable. Not very stretchy but rather flexible. However it is ugly AF and I lost quite a bit of detail. Smoothing caulking by hand is too arduous to get good, consistent results. Using a negative mold with cooking spray as a separation medium would probably get better results - making sure to press the silicone into all the corners and not break the surface between the silicone and the separation medium. I know I promised not to contribute to the cringe silicone faces on the internet, but shouldn't be too bad since I recognize it is cringe and you can hardly tell its supposed to be a face :D Next step is to try a negative mold. Then coloring the silicone.

>>44779 Power mesh fabric—typically an 80-90% nylon and 10-20% spandex blend—is an industry-standard material used to reinforce silicone masks. By embedding this four-way stretch fabric within the silicone, makers can dramatically increase the mask's tear resistance while preserving its ability to stretch and move Also Vaseline (petroleum jelly) works excellently as a mold release agent for silicone, particularly for silicone-to-silicone applications. However, because it is thick, it must be applied in an ultra-thin layer to avoid filling in and obscuring fine textures. The positive mold method is by far superior to negative which you are moving to IMO. You just put it on too thick and sloppy IMO. Fine control can be achieved by putting it on thinly and spread it slowly out and you can press a negative of skin texture into it on later passes. Yes I said passes several passes ideally. Suberdermal color passes, semi transparent mid skin color passes, texture passes, surface skin color passes, etc. Building up detail with each pass in color depth and complexity, wrinkles, imperfections, etc. Doing all of this over a positive mold is ideal. Also the silicone should be clear with color mixed into it and each pass semi transparent . To make silicone look and act like real skin, you need to mimic subsurface scattering—the way light enters the tissue and reflects back out with a fleshy, organic warmth. This hyper-realistic effect relies on layering translucent pigments and microfibers rather than applying a single, opaque color.

>>44779 >>44780 Very interesting posts! I'll look forward to seeing how this turns out in the end. Cheers.

>>44779 What I'm learning, is that smooth skin-like silicone, is very difficult to make. Maybe that's why sex dolls are so expensive for what they are.

>>44780 >real skin I'm trying to stay out of the uncanny valley. There are plenty of "realistic" rubberfaces that look creepy af so I'll stick to stylization. Why have realism when you can have better than realism? How about an elephant with blue eyes? >Put it on too thick and sloppy I want to discover a easy, quick method even a lower-skilled anon can use. >positive mold method is by far superior not really, I lost quite a bit of detail around the nose and lips I could try adding more onto this piece, but I'm gonna print a negative mold and see how that works in comparison. Also with a negative mold reinforcing the back with cloth will probably be easier: I could use the cloth backing to "push" the silicone into the mold. However I will need to take more care with the negative mold: the internal surface will need to be smooth-ish. That will probably involve a small layer height (0.1mm, tripling print time) and using aerosol-based primer to smooth it. I'll also be doing some experimenting with flexi-filaments. Did some a while ago but forgot to take a pic. >>44789 > smooth skin-like silicone, is very difficult to make *laughs in latex Halloween masks*

>>44791 POTD >I want to discover a easy, quick method even a lower-skilled anon can use. I'm so proud of you for this, Mechnomancer! :D According to my own personal experience, even low-IQ art students in today's pozzed, DIE-loving Uni environments can get good results with Smooth-on products. https://www.smooth-on.com/ The issue is that it's an entire products sub-industry focused around the kike-run Hollyjew industry; basically priced out of the reach of typical Anons. I doubt not that the based Chinese are producing more-reasonably priced alternatives, now that they are invading this space in a big way. Bears investigation, IMHO. So (FWIW) I'd probably amend that agenda to: >"easy, quick, inexpensive method..." Cheers.

Not much work, but quickly threw together an absurdly compact mouth mechanism. Jaw opens and teeth move around behind the (future) lips to give a smile or derpy underbite. Will probably mount the entire thing on a second servo so it can twist from side-to-side in that adorkable sidegrin a la lapwing. (see vid) https://youtu.be/F8CFkwrHX4Y?t=38 Maybe even manage to cram a few more servos in there to move the mouth further? Idk have to remember where I put that finished negative mold.

>>44873 POTD This looks great, Mechnomancer. Looking forward to seeing it in action! >Maybe even manage to cram a few more servos in there to move the mouth further? Please try to keep the counts down; if you can find a clever way to do so, then that will help Anons out who follow in your footsteps! > https://redirect.invidious.io/watch?v=F8CFkwrHX4Y That looks good. It's a nice '2.5D' blend of 2D waifu aesthetics, with what appears to be pretty comprehensive 3D facial modelling/rigging. Cheers, Anon.

Ain't been up to much. Mostly enjoying the good weather. Did manage to get SPUD's second leg together and re-worked the torso/support rack so she is being supported by the hips instead of between the shoulderblades. Also re-worked the torso rotation so she will be able to rotate the top half left/right while on the rack. I'll have to get around to re-wiring/powering the servos. Also streamlined the main control program quite a bit. I'll be waiting a while before continuing my pursuit of rubber faces, but I did a little re-work of the kigurumi face: wood filler on some cracks, paint, and new eyelashes made of craft foam.

>>45093 Thanks for the update, Mechnomancer. That sounds like a nice design choice for the torso/support rack improvements. Hopefully this will allow you to put a lot of focus on the pelvic volume as the locus of her major mechanical dynamics. That will be an essential step IMHO for allowing dear SPUD to eventually become fully-autonomous with clear bipedal mobility. Cheers, Anon. :)

>>45093 Looking good! The new face is great! Hopefully in the future, she can walk with the support. I recommend a quick-detach method for the mount.

>>45093 >>45109 >The new face is great! Yeah, @GreerTech is right, the face does look quite good! Cheers, Mechnomancer.

>>45094 SPUD now has all the motors necessary to walk (theoretically(, its just a matter of connecting them and calibrating. >>45109 The pelvis has a quick disconnect pin I took off a 300lb linear actuator. SPUD looks kinda sassy ngl :D Also tends to tilt her head only to 1 side, probably just a programming error for the neck twist.

>>45131 >SPUD now has all the motors necessary to walk We all hope to see her walking soon!! Do you have a mobile gantry solution yet? >SPUD looks kinda sassy ngl :D Heh. Her eyes do look quite good now. Cheers, Mechnomancer. :)

As thread 2 approaches the informal 500 post limit (my, how the time does fly!), brevity of the soul of wit, so let me be brief. After hastily getting SPUD ready for public exhibition I found out the following: - neck motors consumed too much power & had to immobilize them, need a more supportive mechanism - splitting everything into libraries/multiple programs makes it easier to debug & calibrate - using the piper streaming terminal command via OS library in python is more laggy than the original mouthflap queuing program (need to retrofit old programs) - camera for image/object recognition can be hard to aim (dang head motors going on the fritz!) - use metal servo horns as plastic ones wear out fast Public interaction feedback: - need feedback (sound/movement) when SPUD recognizes a voice command (or when she doesn't, "confused eye swirls" might work) - some ambient body movement draws attention to the robit (such as the torso twist) - randomly saying things/waving is funny - the robit choosing *not* to follow a voice command with snark is also funny - folks thought the screen eyeblink was physical In a bid to simply get SPUD mobile (and going out into the world), I'll be putting development of Teslabot-style legs on the back burner and make SPUD something kinda like the attached robit pic, but with the "arm" the torso is mounted bending the other way to be reminiscent of legs. A robit "mermaid" with a mobile base. For ballast, a hefty Uninterrupted Power Supply.

>>45190 POTD Thanks for all the tidbits of information, Mechnomancer! >...and make SPUD something kinda like the attached robit pic >For ballast, a hefty Uninterrupted Power Supply. Excellent choices! I'd recommend you go over @Lin's designs (cf. >>24744, ...). Also, if you find time, some refinements on these concepts came out during dear MaidCom's threads. Please keep us all up to date on your further adventures with dear SPUD. Cheers, Anon.

>>45193 A contribution for robowaifu devs. Is there a thread dedicated only to code snippets and a description of what they do? If not it, it would be a good idea :] Now if you're going to publicly exhibit your work but are worried about your cheeky free-spirited LLM saying naughty words, this code will help: it takes your input string and checks it against a list for any naughty words (form of a text file, 1 naughty word per line plz) and replaces it with asterisks. I whipped it up last night. Some lovely fellows on github created lists of naughty words in multiple languages for all your needs, whether it be censoring or you wanna learn new naughty words. https://github.com/censor-text/profanity-list

>>45197 Naicu. >Is there a thread dedicated only to code snippets and a description of what they do? If not it, it would be a good idea :] No we don't have one. We do have a dedicated programming thread though : (cf. >>37647 ). If you think that's insufficient then yourself or one of @Kiwi or @Greentext anon might make one for us? Please think about what you want to say in the OP text. TIA, Mechnomancer. Cheers.

If this guy can build a smol walking in a few days, I don't see why I couldn't try similar over a few months :D https://www.youtube.com/watch?v=gEjg179fvmc

>>45252 NEAT! This looks like a fun little project. Nice find, Mechnomancer. Do you really plan to follow this approach do you think? The square-cube law will require your actuators to be much bigger/more powerful than the ones this man is using. >"...Including Submarines..." LMAO. Not really the best advertisement IMHO. :D

>>45253 These 80kg servos should do the trick. Meanwhile, I've been hastily putting together the "mermaid" base for SPUD. A pair of parallelograms, with a Uninterrupted Power Supply in the base (as well as some chonky angle iron) to keep her bottom heavy and right-side up. SPUD is now a cohesive, singular unit, and can be plugged into the wall to charge. Given Carry's performance with the battery pack, I expect probably about 4 hours of runtime while unplugged. Perfect for conventions! Weird thing is the ASMC motors are rated for 180kg, yet they don't have the power to move her even though I can move her by hand when unpowered (she sinks when unpowered, too). So I just bypassed the onboard motor and slapped some linear actuators on there. This also means she stays standing while unpowered. I'll probably put a few more panels on there to make her tail look more leg-like. The experience with these ASMCs are making me don my tinfoil hat: the Chinese NERF their exported robit products to try to kneecap the west.

>>45388 It could simply be "The AliExpress Effect" >>45245 For what it's worth, I have had a mostly positive experience with Chinese products. There were some negatives and weird things, but so do Western products

>>45388 Excellent! >80kg servos Those look to be top-quality. May they give you a long and enduring life of service, Mechnomancer. >4hrs That seems pretty good, given dear SPUD's complexity so far. Have you ever used one of those cumulative power meters? It'd give you an accurate measure of the 'Amps Over Time' she's using during a typical outing. I'm planning to get one myself soon for CartWaifu, &tc. >linear actuator I saw that. With this "fishtail" design, I suppose you can just use a single one of them? >I'll probably put a few more panels on there to make her tail look more leg-like. Good idea. >>45389 Good points, GreerTech. And I actually have had good results with Chinese systems as well. Frankly, I think their DJI drones are second to none for consumer use, such as we here on /robowaifu/ have need of.

>>45410 >cumulative power meters The Uninterrupted Power Supply can give runtime and current power consumption. Its a 900 watt/hour pack. >Single actuator I'd have to run one the entire length of the legs, then it wouldn't be very versitile. With a 2 actuator system, depending on which sequence they're activated SPUD can move her center of mass forward/backward on the stand, as well as going into a "sitting" position. >good results with Chinese systems as well It always seems to be that their motor controllers are severely underpowered. I have a pwm controller that's supposed to be rated for 60 Amps and it can't even provide 20 (I know its not a power problem because the big motors work fine with unga bunga relays). Then these ASMCs having underpowered motors as well, I am starting to see a trend. Have to re-print the torso twist servo mount, as it broke while I was moving her thru the workshop. New version will have machine screws threaded into the print to prevent it from snapping like this again.

>>45418 >It always seems to be that their motor controllers are severely underpowered. I have a pwm controller that's supposed to be rated for 60 Amps and it can't even provide 20 (I know its not a power problem because the big motors work fine with unga bunga relays). Maybe it's a "hypothetical" max. Like how I could run a 9B model on my laptop, but it would be incredibly impractical and put strain on it's processors. Or how an ebike/scooter has a "15 mile range", but that's 15 miles of level ground without ever stopping, so it's more like 10 miles. Or maybe it's the voltage that is causing it. Power is made of both voltage and amps. Or you simply got a bad board brand. With enshittification and the rise of Chinese technology, the "haha China low quality" view holds less and less water.

>>45424 >Maybe it's a "hypothetical" max. A motor controller rated for 60 Amps can't provide power to a motor that pulls 15-20 amps (25% of the expected max load). Not a problem with the powersource as the motors work fine when all I did was swap in a dpdt switch wired for polarity reversing. I know for a fact how much current the motors pull as I experimented with different fuses. Also ASMC motors work fine when I connected them to an external linear actuator. >With enshittification and the rise of Chinese technology, the "haha China low quality" view holds less and less water. The phrase "Potemkin village" is foreign to you, isn't it? SPUD is being a little twerp and being reluctant to start up, oftimes getting stuck on the raspi boot screen. Seems to be a problem with the SD card (runs fine with a different SD card, SD card runs fine in other raspi), so I'll probably have to flash another one. To be fair, it is the original one I've been experimenting with raspi AIs with so I've been putting it through the wringer. Also seems to be some sort of glitch when initializing the servos (makes 'em spazz out) so I need to run a special code to initialize the joints before activating the main programs. Had some fun little conversations testing SPUD while doing minor adjustments, which were enriched by the image/object recognition AI running during every LLM prompt so she can "see" the environment. Thought she was in some catacombs so I turned on the light. There still is a bit of lag, most likely because of a) the calculations for her eye graphics and b) running her eyebrow servos off the on-board power (need to re-wire that board to external power) Also had to limit "memory" length as after about 30 exchanges she went a little nutty. Could be the model. To be safe I go with 11.

>>45481 >The phrase "Potemkin village" is foreign to you, isn't it? Sounds like a cope. If it was so shit, tech leaders wouldn't be begging and crying for Congress to ban Chinese cars, phones, and now AI. Here's a hint, it's not an OpenAI 1.7B model you're running. That being said, it does sound like the motor controller was a ripoff. Sorry that happened >Also seems to be some sort of glitch when initializing the servos (makes 'em spazz out) so I need to run a special code to initialize the joints before activating the main programs. Voltage spike?

>>45481 Are the speech and vision models seperate? If so, I may have a solution

>>45482 The Qwen AI team is based out of the Republic of Singapore. The infamous Unitree robo-doogs use a raspberry pi (with the username/pasword combo of "pi/123" The unitree doogs also have a backdoor allowing unfettered access by Chinese companies not only to the robit, but to any device on the same network as the robit-backdoors like this are the reason China tech is getting banned. I also have a Chinese uninterrupted power supply that thinks a 250 watt power converter pulls more than 1000 watts (I know it doesn't pull more than 250 watts because it has overload shutdown). Source on the unitree robo-doogs. https://www.youtube.com/watch?v=lAuvVsZrVMA

>>45485 >The Qwen AI team is based out of the Republic of Singapore. Did not know that. That's good for us politically. >The infamous Unitree robo-doogs use a raspberry pi (with the username/pasword combo of "pi/123" That's arguably the point I want, best of both worlds As for the password, that's the kinda thing an idiot would have on his luggage! >The unitree doogs also have a backdoor allowing unfettered access by Chinese companies not only to the robit, but to any device on the same network as the robit-backdoors like this are the reason China tech is getting banned. In contrast to the totally privacy focused and secure US Tech. And I'm sure cheap competition has nothing to do with it, just like how the OpenAI model just happened to try and hack HuggingFace, the center of Open Source AI, out of all the targets in the internet >I also have a Chinese uninterrupted power supply that thinks a 250 watt power converter pulls more than 1000 watts (I know it doesn't pull more than 250 watts because it has overload shutdown). Genuinely don't know what to say. I believe you, but I have never encountered that problem. Where have you been getting this hardware?

>>45418 >Have to re-print the torso twist servo mount, as it broke while I was moving her thru the workshop. New version will have machine screws threaded into the print to prevent it from snapping like this again. As I've stressed repeatedly across this board, its * the literal nexus of all the mechanical dynamics in a humanoid form-factor. This is also why I urge Anons going after bipedalism with their robowaifus to focus as much mass there as is feasible. That combined bulk-mass then becomes a "free-floating 'cornerstone' " of sorts, which then gives mechanical advantage for all the other constructs (torso & head/arms/legs) while in motion. Otherwise, you're always fighting against the mechanical disadvantage of "thrown weight" (by locating component mass out on the end of those mechanical levers); using up your power & other resources (including wear) more than you otherwise might. Simple as. <---> I'm really glad to see you solving all these issues in turn, and continuing your forward progresses with dear SPUD, Mechnomancer! Cheers. :) --- * 'It' being the lower torso; roughly the equivalent of the pelvis/belly volume.

If you want to increase your LLM speed, you should try LFM2.5 1.2B