Please don't take the threat of Li-Ion explosions/fires lightly, Anon. nitter.net/roller2426/status/1572892393791455232 nitter.net/JimNotBob/status/1456705971993169924 >=== -add 2nd link

News: Yuge Li ore deposit in Burgerland discovered, possibly the largest found on Earth so far. www.science.org/doi/10.1126/sciadv.adh8183 nypost.com/2023/09/11/lithium-deposit-found-in-us-mcdermitt-caldera/ >=== -add 2nd link -minor edit

>>25325 Old news, but thanks for the reminder. IF I use Li-Ion at all, then in something like additional belts as external source. The right way to deal with such a fire seems to be wet sand, btw. Difficult in case of a bus, but it should work with a smaller device.

>>14332 >we typically don't use them as fuel because they're not as energy dense We don't use them cause they cost more per liter than gasoline/diesel and used oil is considered worthless waste; same applies for etanol.

So after years I am first person to point out LiFePO4 (lithium iron phosphate) batteries are safer than lithium ion? They dont heat up as much and are more difficult to explode. They also are better value for the cost in the long term because they do not lose capacity as fast. Part of the reason being they are able to operate lower voltages. But down side is they are heavier and less energy dense. Under development are NaFePO batteries, which likely would be cheaper if they ever come around and are supposed to perform better than current sodium ion batteries unless I had misread.

>>800 >40 days w/o food this is a really interesting phenomenon that i'm not sure will be able to be replicated in machines. my best guess is that the body does some incredible optimization after a while, so you spend energy a lot less than usual

>>30197 that and beer lol, its why monasteries have a reputation for brewing because of lent

>>30199 what does beer have to do with lent? do worshippers go on a beer fast or smth?

>>30200 the monks in medieval europe started it because it didnt count as food so it didnt break your fast, they then got the reputation because they were brewing the strongest beers out of necessity

>>30201 how do you mean "necessity"? does stronger beer help you stave off hunger?

>>30202 Bread was originally invented as a way to store beer. High enough concentration of grains and it becomes liquid bread.

>>30212 >a way to store beer today i found out...

>>30171 >LiFePO4 (lithium iron phosphate) batteries are safer than lithium ion I was mentioning LTOs instead, for years. Maybe LiFePO4 wasn't mentioned explicitly, but sites and channels where these batteries are talked about were certainly mentioned at some point. Or it was assumed people knew about it, or would find out when they'll need to. It might also a cause of going for the extremes, LiFePO4 is in the middle between lithium ion and LTOs. In my case, my argument is the number of loading cycles which speaks for LTOs, which is of course more important for a waifu body where the batteries are hard to access. Yes, maybe in a belt, boots or other external battery holders like mentioned in >>25328 lithium iron phosphate might be a better alternative.

This might be a moot point, but has anyone considered removing the computation from the machine body and treating it like a client-server relationship, where the body sends and recieves signals to the "brain"/server?

>>30249 I've been mentioning LiFePO4 for years. I even discussed charging them.

>>30252 I'll iterate; by having a low power client sending signals to a larger computer that is connected to power, there should not be too much of an issue for battery power; any room the computer would fit inside would be able to be used for space for a battery.

>>30253 I see. I was just meaning in this thread specifically I didn't see it mentioned. >>30249 Looking up LTO and it seems a lot more expensive and doesnt necessarily last as long as LiFeOP4. Why do you think you need direct access to the battery rather than some sort of charging port that can be covered? You only need access if you were hotswapping batteries or the battery is dead and wont charge. If you go hotswapping route charge time doesnt matter.

>>30252 >anyone considered removing the computation from the machine body and treating it like a client-server relationship Yes, this is one of the most important epiphany to have. That the compute doesn't necessarily need to happen in the body, it could be some external server. That said, it would be nice to do as much as possible internally. >>30253 Yeah, I didn't remember the details but wanted to communicate that it's not big news to us. Anyone looking into battery comparisons would've come across it, even if it wasn't mentioned, or even just not in this thread here. >>30261 I guess this thread here was just underused, it would be good to link all mentions of discussions about batteries into it one day. >Looking up LTO and it seems a lot more expensive and doesnt necessarily last as long as LiFeOP4 Last time I checked, it was the variant with the most charging cycles. But I don't follow all the new developments.

>>30252 >>30254 This even has it's own thread >>2956 (which it shouldn't have, a lot of threads should be merged one day)

Kiwi’s Basic Battery Guide! Watts are volts multiplied by amps. For example, a tiny computer using 5 volts at 2 amps is consuming 10 watts. Watt hours is how long a battery can supply power. A 50 watt hour battery would be able to power this tiny computer for 5 hours. In reality, it’s best to assume you’ll get less than this as your watts consumed will vary over time and things like voltage regulators and other aspects of her power system will drain the battery slightly faster than expected. Assuming 80% of the batteries power is usable by your waifu is a safe rule of thumb. Lithium LiIon Lithium Ion cells generally have 100-250 watt hours per kg of mass (higher density can be had, they’re uncommon and expensive). 250 to 700 watt hours per liter of volume. They can be expected to last between 300 to 1,000+ charge cycles before dropping to 80% of initial capacity. From there, expect the usable energy to drop significantly over time. Safe cell voltage ranges between 2.5 to 4.35. Voltage ranges between 3 to 4.1 will allow the cells to last longer. LiPo Lithium Polymer cells are designed to reach higher energy densities. They are closer to the 250 watt hours per kg and 700 watt hours per liter than LiIon batteries of the same cost. They will inflate and potentially explode if discharged below 2.5 volts or charged above 4.4 volts. They are tempting given their low cost and high energy density, they’re too explodey for me. These batteries need to be kept at normal temperatures or they go boom. Costs are reasonable. LiFePO4 Llithium Iron Phosphate cells typically have 90 to 150 watt hours per kg of mass. 200 to ~300 watt hours per liter of volume. Expect them to last ~3,000 charge cycles before dropping to 80% usable energy if you abuse them by over discharging them and charging them with poorly regulated voltage. They will last 4,000 to 10,000+ charge cycles before reaching 80% if you are careful. They tolerate over charging and complete discharging, but will lose capacity faster if done so. Safe voltage range is 2 to 3.6 volts. Only discharging to 2.8 volts will allow them to last hundreds to thousands more cycles. More tolerant to heat and cold, try to keep them at normal temperatures. Usually the lowest cost of the 3. LTO Lithium Titanate cells have an energy density of 30 to 100+ watt hours per kg. 100 to 150 watt hours per liter. They can last between 6,000 to 30,000+ charge cycles before reaching 80%. Safe voltage range of 1.8 to 2.8 volts. I’m unfamiliar with them, but they seem to have the potential to last for well over 30 years before they noticeably degrade in usable capacity. They are safe at extreme temperature ranges of -40c/f to 60c or 140f. The most expensive.

>>30276 i thought they only get swol when they heat up, its something to do with gases being produced which is why spotlights still use lead batteries because the amp required for a spotlight is too high to avoid heat from the draw the holy grail will always be pic related though but im not too into biology to figure out why its so hard to replicate in a machine

>>30276 Thanks, that's good for having an overview in this thread here. >Only discharging to 2.8 volts will allow them to last hundreds to thousands more cycles. More tolerant to heat and cold, try to keep them at normal temperatures. Usually the lowest cost of the 3. This here is the important part I didn't know about. Yes, this very likely makes LiFePO4 even more interesting than LTOs.

>>30252 Yup, that is how SPUD >>30161 works: raspberry pi for the avatar, server is a PC w/ RTX 2070 running an LLM and ai TTS. Data latency via wifi is neligible because the data transmitted per response can be measured in kb (server recieves about 20 bytes of text, then outputs a 120kb wav file), delay mostly comes from my graphics card being smol and taking a while. I'm using some premade batteries. They might be expensive (price has nearly doubled thanks to biden-omics) but sometimes they can be found in Harbor Freight's open box section on discount: folks buying the incorrect battery for their tools, I suspect. Comes with a built-in board to help with battery management and whatnot, but somehow I managed to brick one lol. Still not sure if I can fit it into SPUD's body though, might end up making some chonky megaman-esque boots for her to wear to hold the batteries and lower the center of gravity like the robosapien line does. >>30276 I had some LiPos start to inflate/heat up while charging them. Chucked 'em in a snowbank for a day to make sure they were cool then disposed of 'em.

>>30276 >>30278 I kept coming across videos of people buying LTOs and reporting their grade A battery was only giving performance of grade B so they got ripped off.

This has me thinking. Are soft pack batteries safer than hard pack? Since soft pack cant build up pressure so it cant explode so dangerously if punctured or something. >>30285 > I had some LiPos start to inflate/heat up while charging them. Chucked 'em in a snowbank for a day to make sure they were cool then disposed of 'em. Considering lithium is highly reactive to water that wasn't exactly the best move. In future you need a class D powder extinguisher in case of lithium fires. Though in a pinch you maybe could use a box of sodium bicarbonate sprinkled on it.

>>30289 I forgot to mention a few details like I did seal it in a bag first and the snowbank was in the middle of a field :D Talking with a fellow who did beetleweights battlebots that did use lithium batteries he needed contingency plans for insurance reasons. His plan (which he had to implement once) was putting the beastie in a sturdy metal box until the lithium burned itself out. And that is why I'm sticking with Lead-acid for my non-robowafiu projects. Those break the worst you have is an acid spill (which I've had to deal with before).

>>30290 yeah your friend has the right approach, found out these things are unstoppable bombs, all the oxygen is already there in the uh lithium oxide, once its hot enough to free the oxygen you get a runaway reaction of more heat leading to more oxidation leading to more heat and more oxygen and so on and so forth, all you can do is cover it and wait it out because its not stopping

>>30289 >class D powder extinguisher I would really like to never use that in a domestic place. I hope the carbon dioxide ones are good enough. That said,... >>30295 >all you can do is cover it and wait it out because its not stopping Maybe I'm not well informed yet, but I thought wet sand was the best way to go.

>>30299 CO2 extinguishers are for class B and C fires. Class C being electrical fires but I think that's more like fire from the arcing electricity or flaming cables or something. Class D fires are metal fires including lithium though oddly some extinguishers do not even mention lithium. I am seeing some mixed information online some suggesting certain class C extinguishers might work but I think that's possibly wrong since lithium is self oxidizing if I recall right. Probably best to go physically to a fire station and chat with them for ideas if you have trouble finding info from reliable sources. One thing you can do is avoid full charges and full discharges which will extend the battery life and keep it from getting too hot when charging. Yet then again I actually did find out you can actually buy sodium ion cells even now. Probably not ideal for charge cycles but for testing as long as the voltage is as much as you need they might do fine. They can still explode if punctured or something but they dont burst into flames so it might be useful for just testing maybe?

>>30299 any covering will do, just anything to isolate it from the surroundings, its basically thermite at that point

>>30285 > Comes with a built-in board to help with battery management and whatnot, but somehow I managed to brick one Do you still have it? I have a hoverboard battery that was reading 1 volt, and was able to "restore" it by the method in this video: https://www.youtube.com/watch?v=2Ru28-1yMKE In my case the BMS board was glitched, and though I had the battery on the charger for several days (apparently not charging) the battery read 11 volts, when it was actually fully charged. By doing the short circuit- with a pair of plastic handled scissors- the battery immediately "revived" (the BMS reset) and now works properly.

>>30328 Ive done the short circuit thing on some lithium batteries before. They get extremely hot very quickly and risk exploding into flames. Something to do with the heat breaking up crystals formed in them or something if i recall right. It of course causes damage to the battery but yeah it can work.

>>30328 >Do you still have it? Of course, that little shit was expensive. One of these days I'm going to be dissecting it and replacing/recharging the cells individually.

>>30308 Good news on the battery fire front or lack there of. I was watching some electric sail boat stuff, and they now have solid state batteries. You can drive a nail through them and they won't catch on fire. They have super long life of 12,000 cycles and comparative energy density. I've read before that they can be higher. @7:00 EXCLUSIVE_ Secrets Shared by Marine Battery Manufacturers _ Step 381 They have tested it so it will be product soon. This won't help right away but the fact that they have production tested cells means not so very long they will have more of them. The advantages in lifetime and safety will likely drive the market to them.

>>30348 Wouldnt the right battery size be more like something for an ebike, moped or motorcycle?

>>30348 Yes, solid state batteries are being talked about for quite some time, and they might become very interesting, but it needs to be a real product and available in the right size made for hobbyists and robots in particular. >>30350 >something for an ebike, moped or motorcycle Still too big. At least the later two.

>>30356 its been a thing since forever its just not used as a battery most of the time because its no where near as effective as a chemical battery, dont think there will ever be anything better than liion

So a while ago I did this long calculation based on the energy used by humans to do work and an estimate of what power would be needed by a waifu here, >>13408 So today I'm looking a site where they have an actual humanoid robot built now and they say they will have a 2.25 KWh battery pack. https://www.nextbigfuture.com/2024/08/figure-2-humanoid-bot.html My estimate, 2,500W/h. So yes I'm bragging. I wonder what their robot sells for? I estimated a total cost at the link, and I think it could be cheaper but with all the bells and whistles, materials alone $10,846. The highest cost in my estimates was in muscles $6000 and batteries $1,960. Silicon and miscellaneous was also high at $1,000 which is realistic if you use a good deal of it. It's expensive. Before I have mentioned I thought you could get a waifu for $3,000. I'm not so sure now. It would be really hard. You would have to do some sort of miracle, and I think it can be done, for muscles, and likely it would have limited battery power. The price of batteries has come down and if you limited the waifu to easy work, a lot of sitting around and maybe frequent recharges then you could likely chop the battery cost to 25%, lower the electronics cost to $600 (tough to do and a close call), lose the silicon and miscellaneous for $1,000 and crush it to $100 you get $3,056. But that's no profit. I think it would not be too tough to get a material cost of $4,000 but this doesn't include labor and it also does not include making the machinery, molds, etc. you would need to make these in quantity. I believe at the original $10,000 USD cost you could build these, include labor and make a profit but it would take a long time to get back the labor you spent on design and equipment to build these. My guess is they will sell these bots at over $40,000 to account for all these cost. Maybe as high as $75,000 to get some good return on their money. Now this may seem silly to try and cost these but even if the numbers are wrong they give you some sort of guideline on where you need to trim or be super creative to cut cost. Muscles are the biggest. I don't believe we can cut too much on the electronics or batteries. We have no control over those but muscle cost is something we do so it would seem that effort there would be rewarding. I haven't included a dime for touch sensing because I have no idea how to do this cheaply. It's a difficult problem. So far I assume if something doesn't come up you will have to use position sensing and force feedback from the muscles to tell if something is being touched and how hard. Not very elegant, At all but that's what is "cost effective" right now that I can see. The shin and meat on it is also something that needs to be thought about. I have some ideas but they are not whole yet. Bones I think are not too difficult. Cheapest would be concrete and wire mesh. Also Titebond 3 glue and fiber would likely be fine. Cartilage, HDPE is the cheapest. Used milk bottles would work, melted into shape, with string tendons. Just some thoughts.

>>32819 First of all: 2.25 KWh battery is not 2.25 KW per hour consumption in a regular use case. I mean, it doesn't say the robot will need that amount, unless it's supposed to work for one hour (didn't read the linked article). It could work less time or more than an hour. Also, this is a robot walking around for a while and working. It was never realistic to get that for less than 20k as a fully build robot or for even less than 5k. That said, our waifus would sit around much more, and shouldn't need to carry anything heavy. Maybe water containers in some areas, lifting one to fill some can. Or lifting a toddler a little bit (not walking and carrying it around anytime soon). Maybe a 5kg pet.

>>32819 >I wonder what their robot sells for? My guess at this early stage is that Figure AI is trying to compete directly with Tesla's Optimus [1]. Since Elon's stated 'eventual' price point is about US$20K, then I'd imagine the cost of Figure 02 should eventually settle out on a similar amount. Competition for both of them (and others) with the baste Chinese may quickly drive these cost numbers lower, however [2]. Let us hope so!! :D --- 1. related: https://www.youtube.com/watch?v=jEiHJIqRZok 2. 1Bn Optimus'-related : ( >>32829 ) >=== -fmt, prose edit -add hotlink, crosslink footnotes -retract funpost, apparently I was wrong lol

>>32826 >Elon's stated 'eventual' price point is about US$20K I hadn't heard that. I think he could easily get there if the compute was not too expensive. I'm fairly sure I could build one for 10K, likely less, but I couldn't do the programming. I read Elon's self driving chip cost, if I remember correctly, $35. I think this was from a Jim Keller interview who helped design it. He has two in a Tesla car. You can bet his robot likely has the same processors to cut cost. "...Each Tesla AI chip runs at 2GHz and performs 36 trillion operations per second. That performance is possible because Tesla optimized the chips for self-driving cars and dropped anything more general purpose, said Debjit Das Sarma, another Tesla chip designer and former AMD engineer. For example, the chip handles data recorded as 8-bit integers instead of the 16-bit floating-point numbers more common in AI tasks but that require more power to process. For another, it's got an extremely limited set of instructions it can process. And it's got a gargantuan 32 megabytes of high-speed SRAM memory on the chip, which means it doesn't have to wait around while fetching data from much slower conventional DRAM memory...." https://www.cnet.com/tech/computing/meet-tesla-self-driving-car-computer-and-its-two-ai-brains/ (36 trillion operations per second) damn that's fast We were talking about how likely a lot of stuff could be 8 bit and that seems to be true, or at least for self driving. Robots should be the same. It's going to be tough to find something with that speed and mass of built in memory. All that memory on the same chip is like a cache memory and means it can haul ass on compute with the focused task it has. All these 32 bit chips power are wasted when you only need 8-bits. Comparing two 32 bit and the 8-bit take the same amount of time. Maybe we could cheat by having less RAM and use SSD, "...Currently, the maximum speed for an NVMe PCIe 3.0 (aka Gen 3) SSD is up to 3,500MB per second, while a NVMe PCIe 4.0 (aka Gen 4) SSD can hit up to 7,500MB per second..." NVMe PCIe 3.0 (aka Gen 3) SSD 3,500MB = $128 or so. It's likely for us the sweet spot on cost would be to buy a standard PC motherboard with room for a lot of RAM, NVMe PCIe slot and maybe a mass produced one generation back processor. When I buy motherboards every few years that's what I do. Usually I get like two generations back. So the latest processor might be $1,000 I get something for around $100-120 or so. Any older and cost goes up and the performance is not worth the money unless you are a game freak. And I don't play any games. Of course lately I've been lusting for AI power for making graphics, but it's soooo expensive. What I have now is perfect for what I mostly do but useless for AI work.

>>3282 > 2.25 KWh battery is not 2.25 KW per hour consumption in a regular use case. Yeh, could be 225W for 10 hours. I think they were figuring 20 hours. I found, "...Over an 8-hour work shift, an average, healthy, well-fed and motivated manual laborer may sustain an output of around 75 watts of power...."

>>32848 >I hadn't heard that He mentioned it during his 2024 Tesla shareholder meeting : ( >>31772 ) . >You can bet his robot likely has the same processors to cut cost. It does. They have one in the chest (exactly where I'm urging Anons to locate the 'breadbox' [a cooled Faraday cage+protective frame] for their robowaifus). BTW, this AI "chip" is actually a largish assembly of about 1ft^2 in area (think chip carrier on steroids). You can check out the first Tesla AI Day event to see it. >(36 trillion operations per second) damn that's fast Certainly impressive, but nothing like what's coming in the future! :D >It's going to be tough to find something with that speed and mass of built in memory. We won't need to IMO. We're already on a trajectory to deliver robowaifus at a much-lower cost factor, and using nothing more than commodity, specialized compute hardware suited to amateur DIY'rs (RPis, ESP32s, etc.). We may devise some custom encoders if we can't find suitable ones for things like the knuckles, eyes, etc., but other than that we are looking at COTS stuff+3D-printed physical components. The actuators are the biggest cost-factor, by far (since the software [at least ours] will all be free-as-in-speech & free-as-in-beer). >muh home server My intent is to do everything onboard the robowaifu. Any other approach will preclude smooth-and-fun Walk&Picnic in the Park days with robowaifu -- OBVIOUSLY A VERY-HIGH PRIORITY DESIGN GOAL!! : ( cf. >>32036 ) :DD OTOH, having a detachable-randoseru for extra battery/compute/cooling capacity is a given for my designs during these early years. Also, a home server (-room?) setup is a must if you want to produce private & custom -trained language models, etc., as well. >=== -minor edit

>>32851 >We're already on a trajectory to deliver robowaifus at a much-lower cost factor, and using nothing more than commodity, specialized compute hardware I ran across some astound new micro-controller parts and commented on them in the MC thread here, >>32871 Truly amazing stuff and huge enablers.

Really good news o the battery front. “Natron Energy to build GW-scale sodium-ion battery factory in U.S.The new planned manufacturing facility will produce 24 GW” American company. No rare earths or limited minerals. It uses sodium and Prussian blue , a iron based compound used as blue paint. All cost effective stuff. I think it might be possible to make one of these from scratch. Charges way, way faster. Order of magnitude. Last longer, does not overheat, does not catch on fire. Look at their specs. They don’t compare to Lithium-iron phosphate which has better specs, but theirs are still good. https://natron.energy/our-technology They are aiming for bulk storage first. Industrial stuff. Which makes sense. Less packaging, less cost to start. THis could eventually drive the cost of batteries to very low levels. A rule of thumb is you can mass produce a product for 10% over material cost. A rough but fairly valid rule and the cost of this stuff is next to nothing.

>>32997 Very encouraging, Anon... here's to hoping your prediction comes true! Cheers. :^)

>>32997 A glitch in the sodium prussian blue battery. Not as good Wh/kg. Basically between lead acid and Li-ion. 70Wh/kg. Though this is not really a total disadvantage. Especially if it is cheap. The advantages of it are very high.

>>32997 >>33013 >doesn't catch fire That's all I need to hear. Less energy density is a perfectly acceptable tradeoff for not having to worry about catastrophic battery failure when I'm cuddling mai waifu.

>>33015 >Less energy density is a perfectly acceptable tradeoff for not having to worry about catastrophic battery failure when I'm cuddling mai waifu. THIS.

Related: >>33073 > Post-apocalyptic power supply, EMP robustness, internal solar steam driven energy support