/robowaifu/ - DIY Robot Wives

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

Reports of my death have been greatly overestimiste.

Still trying to get done with some IRL work, but should be able to update some stuff soon.


Max message length: 6144

Drag files to upload or
click here to select them

Maximum 5 files / Maximum size: 20.00 MB


(used to delete files and postings)

Bipedal Robot Locomotion General Robowaifu Technician 09/15/2019 (Sun) 05:57:42 No.237
We need to talk about bipedal locomotion. It's a complicated topic but one that has to be solved if we are ever to have satisfyingly believable robowaifus. There has surely already been a lot of research done on this topic, and we need to start digging and find the info that's out there. There are some projects that have at least partial robolegs solutions working, but none that I know of that look very realistic yet. We likely won't come up with some master-stroke of genius and solve everyone's problems here on /robowaifu/, but we should at least take a whack at it who knows? We certainly can't accomplish anything if we don't try.

I personally believe we should be keeping the weight out of the extremities – including the legs – while other anons think that we should add weight to the feet for balance. What's you're ideas anon? How do we control the gait? How do we adjust for different conditions? What if our robowaifu is carrying things? What about the legs during sex? Should we focus on the maths behind MIP (Mobile Inverted Pendulum), or is there a different approach that would be more straightforward? A mixture? Maybe we can even do weird stuff like reverse-knee legs that so many animals have. Robofaun waifu anyone? What about having something like heelys or bigger wheels in the feet as well?

I'm pretty sure if we just put our heads together and don't stop trying, we'll eventually arrive at least one good general solution to the problem of creating bipedal robot legs.

ITT post good robowaifu legs

>tech diagrams sauce
Open file (316.07 KB 1500x949 poppy-humanoid-github.jpg)
>>4441 Oh, so you're saying it only makes sense to print it out and try it, when one knows about the size of the replacement dc motors? When I mentioned NN here I didn't mean some AI to create a whole waifu mind, but specifically to putting all the states of the motors and joints in, then learning from falling or not falling. Maybe looking not only into the last state, but the ones before which lead to failing or success.
Does anyone know a website with a search function to compare as many electric motors as possible. Didn't find anything and don't have the nerves for it today. I didn't mean AliExpress. Need to filter them for weight and other values. It not only for finding motors for >>4436 (Poppy), but yeah, this would a good example.
>>4473 Based on what this Skyentific guy on Youtube is using for his arms I endet up to find out that drone or RC motors will probably be what's needed to have the same power and then implement the sensors and controls in another way. I found them with "brushless multirotor motor". More I don't need for now. When I have my printer I might play around with it, first only thinking about sensors. >>1602 Link broken, new one: https://www.epfl.ch/labs/biorob/research/humanoid/walkman/
Open file (27.43 KB 450x450 7x20mm.jpeg)
>>4504 One thing you have to be able to accommodate is extremely high RPMs with these motors, especially the tiny little ones for whoop-type micro drones (~50'000 RPM!). They drain energy rather quickly as you might imagine. But, they can impart a remarkable amount of lbs-ft in short order. Think of them kinda like F1 engines for fairybots heh. Example: Apex RC Products 7x20mm Fast 17,900kv / 66,000rpm CW CCW Brushed Motor Set
Open file (37.18 KB 566x415 67020.jpg)
>>4505 Yes, some gearbox for reducing will be necessary. But, you really picked a extreme one, and also I meant rather the flat motors. In the video https://youtu.be/WKRLlthr9kY he is using something like on the pic. I have no data for it, since it seems to be out of production, similar looking ones have a RPM of 1200-5000 which would need to be reduced with a factor of about 150 or 200, I guess. Maybe that not the right approach anyways, the second picture is how the Dynamixel looks like (with ~200 reduction included). I'll look into it on occasion, don't know yet if it will be possible to get a gearbox separately. It's far from urgent anyways.
>>4510 >But, you really picked a extreme one Hehe. That's because I bought a few of these for use in robowaifu arms and to have an 'excuse' to shell out some readies for a whoop drone :^). These little babys are so lightweight, that you could embed 5 of them into the end of each forearm just below the wrist w/o paying to high a lever-moment kinematic penalty. And, given their ability to wind up basically instantly, their ability to impart a remarkably high amount of (accumulated) force in short order makes them a unique possibility for driving strong, lightweight robowaifu hands. Inexpensive too! As with all engineering, everything is a tradeoff. These scalding little bitches put out the heat force of a thousand suns. In their drone use, the motor's shells are both open to air and located directly under their equally high-speed rotating blades, which cools them off. We'd have no such luxury with them inside of our robowaifu's wrists, and I still haven't managed to puzzle out a reasonable cooling system for a 5-set yet. If we can somehow manage that together, then it would be an ideal use-case for these great little motors.
>>4513 I rather wonder how much torque they have, and how you want to add a gearbox or other reduction? However this might fits better into >>406 for actuators in general. We still don't have a thread for hands and forearms or arms in general. I wonder if there has been one in the old forum and will it be imported?
Reminder: This was possible 6 years ago: https://youtu.be/8uiPHL-xey4 and this is from recently: https://youtu.be/aVJs_x2a9z4 I don't think people should focus on walking, but for example if one doesn't need a human-like form like a doll, without making a lot of noise while walking, then developing something might be much easier. Same goes for something rolling with feet like little tanks: https://www.thingiverse.com/thing:972768 (okay, that's a bit OT in a thread about bipedal walking) Gymnastics and dancing: https://youtu.be/BhYUjSAwNB4
>>5223 >okay, that's a bit OT in a thread about bipedal walking ehh, it's not too bad. we do have a wheelchair waifus thread >>2983, so it might be there, but this is probably the best place imo.
This here >>5358 is about walking, James Bruton released his example of a walking bot 2018.
Walking? Meh! Balance bot wheels for feet. "Rollergirl never takes off her skates".
>>5634 you must into pix plox
Open file (162.01 KB 1594x1093 hovershoes.jpg)
Open file (48.07 KB 900x614 new_cubli.jpg)
>>5635 Some things I have been looking at: #1 Hovershoes. Off the shelf solution for locomotion with many open DIY solutions available. Balance control becomes an issue for weight above hip line. Might suggest thinking about internal gyroscopic stabilization. Give it its own gravity well it its chest. #2 the Cubli robot uses 2 motorized disks to spin up manipulating balance. Legs can act as dynamic shocks instead of primary locomotion and balance. Air muscles are viable for this job since the legs need to loosely hold positions instead of supplying quick motion. What used to be a complex circuit for understanding balance and motor control can be done with an opensource RC flight controller with tweaked PIDs. $.02
>>5645 Thanks. Yes I think both hovershoes or some other form of motorized system could serve well as motive 'footwear' in the interim. And gyro-stabilization is also generally a good idea, if power-hungry. Ultimately, typical human bipedal locomotion is the goal, but in the meantime we'll settle for alternative approaches as needs be. >PIDs. So, for the uninitiate, PID is some sort of feedback control mechanism then?
>>5667 Apologies, that page seems to be account-walled. Here's a playlist https://invidious.snopyta.org/playlist?list=PLn8PRpmsu08pQBgjxYFXSsODEF3Jqmm-y
>>5667 Turns out this guy Brian Douglas has his own independent YT channel. https://www.youtube.com/channel/UCq0imsn84ShAe9PBOFnoIrg
>>5671 Apologies, don't mean to bang on about it, but I'm finding more good resources as I dig along. https://engineeringmedia.com/resources
here we have an interesting mechanism https://www.youtube.com/watch?v=uv-Qp8p8Jqg
>>5930 That is interesting thanks. Is there a paper or anything more available on it Anon?
>>5931 nah this was random video in my youtube feed. I am wondering if this mechanism will be enough to achieve balance
>>5933 Here's his channel, seems like he's done a few of these. Amateur designer? https://www.youtube.com/channel/UC7hw6ztc_mnFh630R0MS5uQ
>>5936 https://www.youtube.com/watch?v=vlBgMg9WFfA men god. Luckily my family is heavy into machinery and maybe I will be able to build this. only the mechanism not the propulsion
>>5982 I think if you used 4 of those MIT Mini Cheetah clone thin disc actuators talked about here >>4890 in the hips, then you could probably pull this off. Apparently they run ~$350-400 ea. so plan on about $1500 for the actuators. But if you can get this working it would be both relatively inexpensive and remarkably efficient afaict. That anon's exact design would need some modification. The support frame for the hips would have to separate out the two pairs far enough apart to work well for the onahole cartridge inside robowaifus. https://biomimetics.mit.edu/
>related xpost >>8704
Feet development, related to DARA project >>8704
Are we not making this harder than it needs to be? For example there are reasons humans have a higher center of mass and it takes more power and effort (and years of practice) to walk: our need to be 5 or 6' tall fluid filled bags for one Lets take advantage of the clean slate we have with robo waifus. Several stacking advantages could make mobility a non-issue 1. Unfixed center of mass in design: as in "megaman" robots (and the Roll example), center of mass is drawn to the Boots, which themselves could be weighted further. (for "intimate" times the boots could be taken off to reveal more human-like feet, when sitting comfortably or lying down, etc). The rest of the structure doesn't need much, we don't need steel I beams for a skeleton, and the power battery and heavier parts could be set lower on the frame for the purpose of lowering the center of mass. 2. Gyroscopes: as in the Cube example above, gyros provide stability via angular momentum which could even be used to "push" against to maintain balance. Not sure how power consumptive gyros are but I imagine once they are set spinning it doesn't take a whole lot more to maintain the momentum with decent bearings and lubrication. 3. AI assisted learning. Same as how we learned to walk. Trial and error until a way to tackle moving from any point A to B across any slope or terrain is simply reflex. (I'd like to talk more about assigning movements to "reflex" CPUs, this is key since obviously we dont go thinking about every specific movement we need to make, we just think "go down stairs" or "open the door" and our habit and reflex handles the rest, correct?) Probably more to add to this list, but I see no reason these three advantages would not stack in a way to solve the problem of mobility simply and elegantly.
>>9053 The idea has been bounced around here of a Roll-like waifu, and certainly keeping track of center-of-gravity is vital for bipedal locomotion. As to your third point, particularly > (I'd like to talk more about assigning movements to "reflex" CPUs, this is key since obviously we dont go thinking about every specific movement we need to make, we just think "go down stairs" or "open the door" and our habit and reflex handles the rest, correct?) This seems to be pretty closely related to the idea of 'Neuromorphic Computing'. (You can find a paper in the Library thread). I'm inclined to agree with your ideas on this, and it's certainly good biomimetics, nervous system-wise. It entails it's own complexities too ofc, but I imagine the tradeoffs will be well worth it. Good ideas, thanks! It would be nice if you'd sketch some ideas out for us so we can see clearly what you mean about 'stacking advantages'?
>>9053 Some of this is food for thought. However, we're going to see different approaches, because people have different priorities. For example: I wouldn't want her legs to feel to light when I lift them, also her not needing to wear boots all the time. For outside, it would be okay though. Also, I don't get the problem until I tried it out myself. Humans do it, so it should be solvable. Another thing is, bipedal walking without help or guidance (walls etc) should be towards the end of any priority list, imho. For a domestic girlfriend it isn't very important. She could walk with help, using walls, being driven or drive herself with some device, walk on all fours, dancing while gripping something (e.g. a pole), standing up after grabbing something, then pose or do the dishes while keeping balance,...
>>9054 >Good ideas, thanks! It would be nice if you'd sketch some ideas out for us so we can see clearly what you mean about 'stacking advantages'? stacking as in multiplying, not merely Adding, each advantage Gyroscope + Low Center of Mass is an order more advantageous than Gyroscope alone or Low Center of Mass alone Because each is a different "vector" and not merely adding to the same vector Does that clarify things?
>>9055 noted, but also consider than a powerful enough internal gyro or set of gyros could effectively be an "internal" balance pole of sorts
>>9057 >Does that clarify things? Yes, it does thanks.
>>9672 Thanks kindly for the links, Anon.
>>9058 Agreed. IMO, any practical robowaifu walking system today -- given our current rudimentary state of kinematics sensing, planning, and animation -- will necessarily have some kind of gyroscopic stabilization to succeed. Here's an example of what good gyro systems can do >>5645 > the Cubli robot https://robohub.org/swiss-robots-cubli-a-cube-that-can-jump-up-balance-and-walk-across-your-desk/
Finally someone did a study on minimizing energy consumption in gaits. >We focus on the problem of developing efficient controllers for quadrupedal robots. Animals can actively switch gaits at different speeds to lower their energy consumption. >In this paper, we devise a hierarchical learning framework, in which distinctive locomotion gaits and natural gait transitions emerge automatically with a simple reward of energy minimization. >We use reinforcement learning to train a high-level gait policy that specifies the contact schedules of each foot, while the low-level Model Predictive Controller (MPC) optimizes the motor torques so that the robot can walk at a desired velocity using that gait pattern. >We test our learning framework on a quadruped robot and demonstrate automatic gait transitions, from walking to trotting and to fly-trotting, as the robot increases its speed up to 2.5m/s (5 body lengths/s). We show that the learned hierarchical controller consumes much less energy across a wide range of locomotion speed than baseline controllers. https://arxiv.org/pdf/2104.04644.pdf This could be adapted to bipedal walking by creating different expert policies for not only different speeds but also carrying items, moving hands and arms around, wearing a backpack, different size weights on the chest, cat girl tails and so on.
Adversarial Motion Priors for Stylized Physics-Based Character Control >Synthesizing graceful and life-like behaviors for physically simulated characters has been a fundamental challenge in computer animation. Data-driven methods that leverage motion tracking are a prominent class of techniques for producing high fidelity motions for a wide range of behaviors. However, the effectiveness of these tracking-based methods often hinges on carefully designed objective functions, and when applied to large and diverse motion datasets, these methods require significant additional machinery to select the appropriate motion for the character to track in a given scenario. >In this work, we propose to obviate the need to manually design imitation objectives and mechanisms for motion selection by utilizing a fully automated approach based on adversarial imitation learning. High-level task objectives that the character should perform can be specified by relatively simple reward functions, while the low-level style of the character's behaviors can be specified by a dataset of unstructured motion clips, without any explicit clip selection or sequencing. These motion clips are used to train an adversarial motion prior, which specifies style-rewards for training the character through reinforcement learning (RL). The adversarial RL procedure automatically selects which motion to perform, dynamically interpolating and generalizing from the dataset. >Our system produces high-quality motions that are comparable to those achieved by state-of-the-art tracking-based techniques, while also being able to easily accommodate large datasets of unstructured motion clips. Composition of disparate skills emerges automatically from the motion prior, without requiring a high-level motion planner or other task-specific annotations of the motion clips. We demonstrate the effectiveness of our framework on a diverse cast of complex simulated characters and a challenging suite of motor control tasks. https://xbpeng.github.io/projects/AMP/ Basically a way to use adversarial learning to imitate reference motions while seeking to solve a goal.
>>10160 >Finally someone did a study on minimizing energy consumption in gaits. Excellent. I consider that topic to be essential for finally achieving the idealized goal of a human-like robowaifu without having a state-sponsored R&D budget available. This will only just be on the very edge of what's even technically feasible at the moment.. If it's achieved at all, it will be due to everything on board the robowaifu being designed to be as efficient and economic as possible. Macro physical behaviors like walking will consume significant energy resources. So, not only will efficient gaiting be needed to even work in a natural lifelike way, but it should also make it possible to have our robowaifus walk around for more than 2-3 minutes without immediately needing to do a full 8-hour deep recharge before she can move for another 2-3 minutes, etc., etc. Ever wonder why a humanoid robot that's as well-funded as even Honda's ASIMO is never shown walking for more than just a few steps in any promotional? Because it's gait is just about as inefficient as it could be, power-consumption wise. And even significant steps forward in this arena such as Boston Dynamics parkour bot aren't shown going on long walks, in their case it's more a question of not keeping weights to a minimum, even though the gaiting algorithms are much better in that case. Every.thing. on board must be optimized for robowaifus to succeed and become the reality we all dream of. Everything.
>>10164 I think in real use robowaifus will be more like laptops, plugged in most of the time but able to disconnect and go at any time. My robowaifu would most likely be sitting 95% of the time and my place is so small she could get around while still being connected to power. It'll be an interesting problem navigating small spaces without knocking anything over and not burning excessive amounts of power to avoid stuff.
>>10181 >It'll be an interesting problem navigating small spaces without knocking anything over Yes, that definitely is going to require us to have an effective solution for at least two needs Anon: 1. Body-awareness. Distinct from Theory of Mind, this needs to encompass not only a detailed knowledge of her own physical volume & shape, but also a predictive methodology that can plan for her own kinematic/physics dynamics upcoming, based on her own short-term planned movements. 2. Excellent situational-awareness. She needs to have a detailed map of her environment at all times -- and one that can more or less instantly adapt to changes in it. Your own motions in the space for example, or your cat's. On a related point, until and unless we can actually give her a Theory of Mind, she'll have to rely on situational awareness (and possibly some types of crude heuristics) to be able figure out where you are/shortly will be. Thus (just one of the many) need(s) for a near-instantaneous observation capability.
>>10183 It would make sense to do this in a physics simulation. These were not only meant to train them, but to have one all the time for having awareness. Of course not simulating everything all the time, but when necessary. Also with simplified objects.
>>10186 Yes, that might be a good approach Anon, and for both of the enumerated needs too.
DrGuero made a new short video https://youtu.be/wxH3vQOz3JA, where he mentioned that his code is opensource now. His website http://ai2001.ifdef.jp is a bit slow right now, mb to much load after releasing the video. Also it's in Japanese, so some translation software will be necessary.
>>10343 Thanks Anon! Here's teh code page, I'm DLing now. http://ai2001.ifdef.jp/uvc/code.html >update OK, I've had a quick look at the code. It's old-style 'C with classes' (c.1985 C++) code. It seems to have only one major external dependency, Open Dynamics Engine. https://ode.org/ It's also W*ndows software. I'll try to sort out getting it to build on a modern C++ compiler on Linux sometime shortly, probably this coming week. Thanks Anon, he may be able to help our Bipedal Locomotion progress along.
>>10343 >>10344 update: OK, I've assembled the 4 code files into a workable project. I haven't set about trying to build it yet b/c dependencies. https://files.catbox.moe/wn95n1.7z However, if anyone would care to work on translating the embedded comments in the codefiles, then that would be a big help to me later on getting the code to run for us. TIA.
>>10344 >https://ode.org/ update By all appearances, the original ODE project has been abandoned. However, it's apparently continued under a new team. The project is now located at: https://bitbucket.org/odedevs/ode/src/master/ pamac info ode Name : ode Version : 0.16.1-1 Description : High performance library for simulating rigid body dynamics URL : https://bitbucket.org/odedevs/ode/ Licenses : BSD LGPL Repository : community Installed Size : 1.7 MB Packager : Antonio Rojas <arojas@archlinux.org> Build Date : 19/03/20 Install Date : 10/05/21 Install Reason : Explicitly installed Signatures : Yes I've gotten all the demos to build on Linux, so that's a good sign. I'll have to sort out all the dependency issues in the build file, but I'd think this project should be doable on Linux in the end. As is commonplace for me, I overestimated my abilities and it will be longer than I planned on. I'll post here when I have it all working. Probably be a week or two hopefully.
>>10409 Thanks, I would have helped already, but I have problems with my browser. Also there are a lot of comments, so I'll need to use sed or something like it. Great if you want to work on it, but bipedal walking isn't something very pressing. So don't worry about how long it takes.
>>10410 Thanks Anon, appreciated. I think r/n it's more of a thing to get working early, so that the AI geniuses here will have a readily-extensible tool they can use to work on connecting their AIs into, and coordinating a robowaifu's body movements, etc., via an early simulator environment. My personal goal for the software is to be able to continue refining it into an actual robowaifu runtime software, and for it to work onboard with small SBCs inside her. In my early tests thus far with ODE, it's looking pretty encouraging (~ sub-millisecond collision responses, etc.)
>related xpost (>>10601, pdf embed)
>>5409 Here the whole history of James Brutons bipedal walking robots, back to 2004 to around 2018. Robot X, shown in >>5409, was the end result. https://youtu.be/JWvH5PHKK74

Report/Delete/Moderation Forms

Captcha (required for reports)

no cookies?