/robowaifu/ - DIY Robot Wives

Advancing robotics to a point where anime catgrill meidos in tiny miniskirts are a reality.

Build Back Better

More updates on the way. -r

Max message length: 6144

Drag files to upload or
click here to select them

Maximum 5 files / Maximum size: 20.00 MB

More

(used to delete files and postings)


Have a nice day, Anon!


Computational Parallelism, Photonic ICs, Soft Robotics, and The Use of Low Power Analog Systems Robowaifu Technician 01/09/2024 (Tue) 19:32:19 No.28223
If we were to be realistic about creating robot wives capable of forming seemingly genuine human actions such as fine motor skills, emotion, and rational thought while resembling a human with a few non-human features, i think the best way to approach this is to embrace the usage of soft robotics, computational parallelism, photonic integrated circuits, and low power analog systems. Computational Parallelism: Divide and conquer algorithms are often used to solve computational problems by dividing tasks up into smaller problems, solving them individually, and combining the results together to form one big output. This generates faster results and is already possible to achieve with conventional hardware, but most conventional hardware is limited to the amount of cores needed to effectively solve problems this way on a larger scale. It's a big hindrance to the speed of operations like this, so we'll need something more powerful. There are ASICs available on the market that were engineered to be used this way. However, they're too expensive, which drives me to the possibility of using Photonic integrated circuits. Photonic ICs: Photonic IC are exactly as the name implies: Integrated Circuits that make the use of Photons to computational operations as opposed to electrons. Manufacturing Photonic ICs is relatively more cost effective than if you were to try manufacturing conventional silicon ICs. Another huge advantage is that we can solve computational problems much faster with less power using photons without generating heat the same way conventional ICs do with electrons. The biggest caveat here is that this is still being researched and that special tooling would need to be made for us to build these in house. Soft Robotics: Soft robotics are the best option for creating human-like bodies for robots. Even better is that they often don't require too much power to actuate joints or muscles. Most soft materials use for soft robots are cost effective enough to be made with commonly found materials. I personally think making use of alginate hydrogel would be a smart idea for use in robots waifus because it is hydrophilic, and anti microbial. It is commonly used in medical applications such as wound dressings and drug delivery systems. It is also non-toxic which makes it a perfect candidate for use in robot waifus. I'd also entertain the idea of using other soft materials in conjunction with it for better results. Primarily, it can be used in forming fat-like deposits, skin layers, and artificial internal organs for robot waifus. I think it may also be possible for us to create internal lubrication systems for it due to the fact that it is primarily water based. Low Power Analog Systems: Low power analog systems are vague in definition, but the general idea here is that we can use things such as fiber optic wires, digital-to-analog converter chips, and digitizers to create a type of artificial nervous system that communicates with the rest of the hardware running signal transmitters over echo state networks. It's hard for me to accurately describe my visualization of this, but by now, I'm sure you get the point already. The only problem left to solve is power. How to we obtain it, and how much can we use? I don't know, but if we take the low power-high tech approach, it probably wouldn't become too much of an issue. I think that maybe the best way would be to use electrochemcial energy in some type of internal fuel cell to generate power. All of the systems described above would require less power than if we tried to use servos and silicon chips. I's also say that maybe we could take more inspiration from nature to create bio-mimicking features that we would find desirable. A lot of this is speculation that i got from dipping my toes into various topics and picking out what i think could be useful to finally end the hypergamy going on in the west. This is largely a cultural issue that is mulifaceted in every way and our efforts could be in vain if the feminists and roastoids find a way to outlaw open source tech. Not only should we be focusing on building robot wives to flips the tables in the dating and sexual market, but to also consider political and economic alternatives that absolutely strip and decentralize power from the hands of the elite. I personally think agorism is the way to go but that is a another topic for another time. Let me know what you guys think. Am i an idiot, or am i onto something?
Anon im going to be honest with you. I don't think you can make photonic integrated circuits, I also don't think AI anons can even make a LORA for stable diffusion. Why should i think otherwise talk is cheap. Oh you're too busy. Busy doing what? do you work on the weekends too? Why are you here?
>>28224 Lol, i dunno. I'm just spitballing for the guys fucking around with arduinos and breadboards. Why are you here? To mope and complain? I think if we had just a little bit more autism, we could probably pull a fraction of this stuff off before we lose steam at best.
>>28225 I'm making progress. I am going to make a mask soon that can move its mouth, then I'll make robot hand and make it grab a banana or something based on a dildo picture i showed it lol. I already made a robot hand that kind of doesn't work and talked to an orange pi among other things.
>>28226 That's nice. I'm working on my patience. It's been a while since I've flexed my software development muscles and I'm currently trying to improve my algo design skills right now. I primarily use C but I'm eyeing Nim too. Since i already dipped my toes into various topics, i can infer on other things for reference to re-inforce my understanding of other problems which allows me to be twice as useful. Information theory for example is great for understanding data encoding over serial networks or for audio and video encoding/decoding. I guess we all just have to carmack it out, if you know what i mean. Also, here's some extra coombait.
>>28227 If you're interested in AI that aloha robot project does look interesting. However its meant for their robot. We don't have a robot however we could use the inmoov as a reference or something. Just a suggestion.
alginate begins to degrade and fall apart within just a few hours of it setting up. it cannot be used for a robot beyond a body cast or w/e and a pour done immediately following the cast
when you mention creating ICs, it is a little triggering for me. huge companies have to spend millions and millions of dollars to create ICs and to suggest DIYers in their basement roll their own ICs made on a DIY budget is like shooting us in the foot. We simply cannot do it nor should we. I am all for trying to make custom circuits and custom low level programming but to start fabricating our own ICs is just going too far for a DIYer. Also, you seem to suggest we fabricate our own power storage "fuel cells" instead of using batteries that are commercially available. That triggers me too. That is not feasible nor practical. We have to use tools readily available, mass produced, and affordable for this. To try to roll our own ICs or invent our own batteries for our robots is a recipe for going down a 30 year long rabbit hole and never even being able to start work on the robot at all.
>>28234 I don't see why you'd get mad at that and not the fact that the project has to stall until someone develops chatgpt robowaifu edition and talk endlessly about how that's what matters.
>>28236 prioritizing ai vs body doesn't matter - it is just preference. that is not the same as suggesting a massive reinventing of the wheel on ICs and batteries which just isn't realistic to do solo in a basement with diy funding period
>>28245 remaking chatgpt is a massive reinventing of the wheel technically
Anons, we have inmoov, we have poppy, we have sophie devs files, we have language models, we may even have ai for movements which in my option might be overkill, we have had visual recognition for a while now. We couldn't ask for more resources than what we have now and yet you guys are trying to turn this thing into... i don't know. Look at op's post. Wtf is he talking about.
>>28250 something about hpc and asic which would would solve the problem of ridiculous power usage from using general purpose hardware as an adhoc solution but unless someone actually has contacts with a foundry thats beyond our reach its like with the crypto farce doing the stupid hash with gpus was just an adhoc solution it wasnt optimal or even good, it took years before anyone made asic rigs to do that stupid hash but once they did the power usage was just fractions of using a gpu
>>28246 chatgpt cannot be run on a local machine nor trained by any of us on local machines. it is closed source too. large language models are not diy friendly period. so no, making ai for our robots is not reinventing the wheel in any way. it is a necessity and unavoidable.
>>28246 chatgpt doesn't know a thing about me nor remember a thing we talked about after me interacting with it for a year. doesn't even know I exist. this is one example of why that type of model is not suitable for a waifu or robot friend which would need to remember most everything and build on that interaction
>>28246 also if I convince chatgpt such and such is true in a long argument with it, I can come back tomorrow and it is back to saying it isn't true. It did not update its knowledge based on my convincing it nor does it now take this new position with all other users. so it is sandboxed to your session and you cannot affect it globally. that type of thing is completely unacceptable for our use case for robot friends
>>28257 There are people who are quiet happy and having a blast talking to llama2 which is open source I think. I'll let facebook worry about that. I personally don't care. And if anybody here cared they'd be learning tensorflow or linear algebra or whatever it is that it is required to make the thing.
>>28261 fair enough but to my knowledge there aren't any that remember you and your past conversations with it and learn from those and can be taught things and remember them across sessions and grow to know you more over time and even predict things about you etc. that is the issue and those are must for having a semi realistic relationship
>>28264 I think current llm's maybe capable of that right?
Open file (26.48 KB 815x619 memorysauce.png)
>>28264 LLMs remember your current conversation by injecting the entire history into part of the current prompt. So on the raw end you get something like "user message: I am hungry, history[user: hi, AI: how are you today?]" It wouldn't be too hard to periodically have the LLM pause, sum up a conversation history and add it to the prompt (after all, we do that naturally while awake and also when we sleep). Mostly variable juggling. Similarly, you could monitor chat for certain keywords from the user (such as "my favorite item is x) then save it to an external text file, to be injected into the prompt whenever any word is detected (or just all the time idk) Attached is a vid and some code (I forget if it is complete or not) demonstrating "memory storage" in terms of a favorite book without injecting it into an LLM prompt. As I recall this was actually set up to pull up a preference file based on the individual face detected (up to 5 faces "memorized") using the face_recognition python library with individual face encodings, but since then the face recognition library stopped working cuz I fiddled with something I wasn't supposed to, but the idea is still there. I'm fine letting this be public right now just cuz it is basic variable juggling/modified copypasta so nothing too fancy and I'm not backing it as something 100% working, just something for y'all to try out/modify as you'd like :)
>>28270 well I'm sure the prompt is character limited and that limits the llm to recent discussion and it will not have a exhaustive ability to remember YEARS worth of thousands of conversations where you get to know eachother and so it would be like 50 first dates the movie. terrible.
>>28270 is this seriously how its done, why cant you just dump the memory contents to a file like a save state or hiberfile, reprocessing all the input from a clean state each time seems pretty braindead when its already been done and already in memory like just save it lmao the state not the inputs
>>28290 Do you guys really want a robot that behaves like a human? Like Lt.data from startrek? I don't lol
>>28294 I sure do! I'm surprised you don't
>>28290 Even people don't remember exact conversations, just the important bits (you like X, I like y). And quite a bit of the time, prompt length is limited by your available processing power before the limits of the LLM. I'm sure you could add supplementary context files on various subjects such as favorite foods, favorite drinks, trips, etc that you can load into the LLM prompt depending on the context (aka keywords) provided by the user, and even come up with a dynamic way of making them (heck there is this one AI model that natively searches supplementary docs already for relevant data, I forget the name). In this manner, not only would you have a sort of memory backup of the robowaifu personality but it could even be transferred across models.
>>28294 >Do you guys really want a robot that behaves like a human? Yeah well, that's exactly why i made this thread. There's room here for all sorts of people in different camps. There are people that want to have sex with a toaster, people that want half-mecha half-furries, people that want half-human half-mecha, and normal people that just want a robot that is as close to human flesh and cognition as possible. I'd rather face the uncanny valley head on than to make a cheap automated appliance that i could easily get bored of in a few days to weeks. We may not be able to manufacture special tooling for soft bipedal robots or the hardware necessary for it to process information efficiently as of now, but that shouldn't really stop us from trying. I guess the hard part about being optimistic is that you still have to work within the confines of what is feasible both financially and physically. There's still much to research about ourselves as humans, so we may never truly imprint a natural human frame of mind onto whatever we create. My only goal here is to just have a companion that I love to look at, be intimate with and don't have to worry about so much. If that's completely out of the realm of possibilites (which it currently is), then I'm content with staying alone. I'm not a regular on alogs and my efforts to improve my programming skills are going toward other things that can flip the ongoing culture war on its head. But sometimes it's nice to entertain the idea of picrel serving me and catering to my every needs. I have great ideas that i'd like to try out to make this thing a reality, but it'll have to wait. I'll revisit this little project of mine some other time.
>>28329 Anon if i wanted to have sex with a toaster id buy a sex doll lol The point is that the toaster sexes you Anybody who doesnt to want sex with the waifu raise your hand
>>28330 Sounds like you don't understand what a waifu is, anon.
>>28332 I wouldnt have married my wife if she had said that there will never be sex i know that much
>>28333 This could be a replacement for sex it could not. In my mind the talking is like for the most part nice but not really necessary. Notice how throughout the show chobits she just says chi. Keep in mind that people can get desperate for sex however nobody is desperate for love. Roasties use the desperation for sex as leverage.
>>28333 So... that's a "no". And apparently you haven't read Chobits either.
>>28334 >nobody is desperate for love Brah, you should get your priorities in order.

Report/Delete/Moderation Forms
Delete
Report