/robowaifu/ - DIY Robot Wives

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

Happy New Year!

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)


“Perseverance, secret of all triumphs.” -t. Victor Hugo


Realworld Interaction GreerTech 05/08/2026 (Fri) 02:10:29 No.44457
aka "Does she know how to make a grilled cheese" One area that still needs to be researched for fully-capable robowaifus is the manipulation of objects and the knowledge/ability of how to do those tasks in a relatively random and uncontrolled environment. Hardware is part of it, but mostly what's needed is software. Consider this, if you make perfect robot arms for your robowaifu, what can you actually do with them? Here's a good example of where we are now in the DIY space. Complex tasks require full control, but simplistic toddler-esque tasks can be done by AI. https://xlerobot.readthedocs.io/en/latest/demos/index.html Related threads Vision >>97 Hands >>417 >>4577 >=== -edit subj
Edited last time by Chobitsu on 05/08/2026 (Fri) 19:47:57.
Woohoo! Nice thread, Anon. This is a topic we've discussed broadly across the board, but having it all consolidated together is definitely a needed benefit here. GG. --- BTW, any chance you'd be alright w/ changing the thread subject to "Realworld Interaction" instead? (We can change that.) If not that's OK too, I'll leave it to you.
Edited last time by Chobitsu on 05/08/2026 (Fri) 12:06:33.
>>44459 >BTW, any chance you'd be alright w/ changing the thread subject to "Realworld Interaction" instead? (We can change that.) Yes, that would be much better
>>44462 Done.
Posting ITT for interest since many ways of interacting with the real world might turn out to be applicable to holowaifus. I don't really have anything neat to contribute to this, but I'll watch this thread.
>>44464 >since many ways of interacting with the real world might turn out to be applicable to holowaifus. Yes, absolutely HoloAnon. Most of the algorithmic control systems will be just the same whether it's virtual interaction or realworld interaction.
While /robowaifu/ was on hiatus, I found out what may be the special sauce for this problem, Vision Language Models. https://en.wikipedia.org/wiki/Vision-language-action_model While this is clearly a good path to follow, it's important to know that it isn't "finished". It takes a lot of computing power to train these models. Here's a good resource for open-source VLAs https://huggingface.co/docs/lerobot/index
There is also something slightly less advanced, called a Motion Diffuser Model. Basically, it turns text into actions. It's like if someone told you to do an action, and you do the action without having to be told exactly what to do. https://github.com/GuyTevet/motion-diffusion-model This will be great, because you can easily use an RP model with this. Unfortunately, it would be almost like scripted actions.
>>44521 *Vision Language Action models
>>44521 >>44525 >>44526 POTD Great work, GreerTech. Cheers. :)
Thread related video >>44632
>>44635 Yeah, I should have linked the video in this thread instead of that one. I liked the part near the end where he said AI needs to predict what happens soon to compensate for it's reaction time, the same way a human brain does. And I mentioned in the comments that a problem with this is finding the right balance between rewarding an accurate prediction of the future without just letting it stare at a wall for a reward, while still being useful for boring, repetitive tasks that you wouldn't want to do yourself.

Report/Delete/Moderation Forms
Delete
Report