Sourccey, another hint that now is the time for home robots

[Alan Timm]

Could 2026 actually be the year of the home robot?

This R2D2-esque 3d printed robot seems to think so.

Everything seems to be coming together right now.

https://sourccey.com/

https://www.youtube.com/watch?v=6QjN8-lxYSU

[Carl]

Looks like we will all be spending the next 5 years training our robots :-)

[Alan Timm]

Sourccey being taught how to fold laundry.

Nick and his brother had their ups and downs, but great progress nonetheless.

https://www.youtube.com/watch?v=g1hKfgJLgCc

[Thomas Messerschmidt]

Alan, 

I hate to be a “Debbie downer,” but every robot since Topo in 1983 thought it would usher in the year of the home robot. The only robot that has come close was the roomba. Still I do remain hopeful! 

Ok, this one is learning how to fold laundry, but I thought that was a skill that the PR2 could do 15 years ago. How is this different? Can’t the training be transferred to a new robot? (Obviously I know very little about this.)

Thomas Messerschmidt

-  

Need something prototyped, built or coded? I’ve been building prototypes for companies for 15 years. I am now incorporating generative AI into products.

Contact me directly or through LinkedIn:   

https://www.linkedin.com/in/ai-robotics/

On Jul 30, 2026, at 9:06 PM, Alan Timm <gest...@gmail.com> wrote:

Sourccey being taught how to fold laundry.

Nick and his brother had their ups and downs, but great progress nonetheless.

https://www.youtube.com/watch?v=g1hKfgJLgCc

<screenshot_20260730_210535.jpg>

On Monday, December 22, 2025 at 1:04:38 PM UTC-8 Carl wrote:

Looks like we will all be spending the next 5 years training our robots :-)

On Sunday, December 21, 2025 at 1:24:29 PM UTC-8 Alan Timm wrote:

Could 2026 actually be the year of the home robot?

This R2D2-esque 3d printed robot seems to think so.

Everything seems to be coming together right now.

https://sourccey.com/

https://www.youtube.com/watch?v=6QjN8-lxYSU

--
You received this message because you are subscribed to the Google Groups "RSSC-List" group.
To unsubscribe from this group and stop receiving emails from it, send an email to rssc-list+...@googlegroups.com.
To view this discussion visit https://groups.google.com/d/msgid/rssc-list/5d452069-32c6-4b86-95af-b447ddecc0c1n%40googlegroups.com.

<screenshot_20260730_210535.jpg>

[Chris Albertson]

On Jul 30, 2026, at 10:42 PM, Thomas Messerschmidt <thomas...@gmail.com> wrote:

Ok, this one is learning how to fold laundry, but I thought that was a skill that the PR2 could do 15 years ago. How is this different? Can’t the training be transferred to a new robot? (Obviously I know very little about this.)

Can the training be transferred?   What I call “motor skills” must be trained on a simulated robot whose specs are exactly like the real robot.     But of course, once you have the model trained, it can be copied onto numerous identical robots.

But abstract facts can be transferred.   The problem is the robots you see today don’t know any abstract facts.  So nothing is transferable.

There is a newer, few-year-old kind of model now that looks like it can learn what they call “intuitive physics,” meaning that things fall if they are dropped, objects continue to exist even if I can’t see them.  Liquid can be stored in an upright container, You can pull a rope but not push it, and 1.000,000 other basic facts.   Then after such basics, the skills like folding towels or using a screwdriver can be trained in less than 200 hours.  That is astonishingly fast.   And then of course, those skills transfer to identical robots.    This helps because the motor skills are learned so fast if intuitive physics is already trained.    But here is the good part: the model that knows this intuitive physics is open source; go and get v-jepa 2.1 at hugging face.    So we have some reason for optimism.

We humans are the same way.   We can teach a child to do long division or place names in alphabetical order but we can NOT tell a child how to ride a bicycle, she has to learn that herself.  We can only teach intellectual activities.  It is the same with training robots.

But even with this, it is not enough.   Today’s robot is trained to fold a towel only if the towel is lying flat on a table, and you line up the robot and then click the “fold towel” button.      What is needed is a robot that notices the drier finished its cycle, checks if the stuff inside is really dry, and folds and puts away whatever is in there, and you never have to tell it to do this.   That is a level of AI we just do not have today.

The root of the problem of why we can’t reuse the PR2 skill on a new robot is that the skill was controlling electrical current to a set of motors.  The PR2 never at any time understood the concept of “folding” and never knew what a towel was.  If the robot did understand towels, then we could transfer this to the new robot.  The skill of controlling motors is not transferable to a different set of motors.   This is what the v-jepa 2.1 tries to fix; it learns what a towel is and what folding means. All of that should be transferable.  But this is new, not yet common. Almost every towel folding robot today is a mindless zombie that does not know the concept of “towel” or “fold.”

[Alan Timm]

Hey Thomas, 

I pulled up the PR2 video you were referencing, and that demo was a seismic game changer at the time.

https://www.youtube.com/watch?v=caVo6EJFV8s

A $250,000 (estimated?) robot that could fold a towel autonomously under strict laboratory conditions in 6 minutes was nothing short of amazing!

Now look how far we have come in 15 years and a handful of technology advances.

A couple of kids built a robot from off-the-shelf parts and trained it to fold a towel in 10 seconds.

And the same training framework is training the robot how to fold t-shirts and whatever else they want to throw at it.

Not against a green backdrop but in the real world, and in real time.

I'd call that progress.  :-)

Alan

[Chris Albertson]

I have to agree.   Some of this is simply advances in computer hardware.   Computers got faster.    But most of the improvement came from a better understanding of the theory.   What was once cutting edge is now taught in universities.    That is the way engineering goes.

Someone here complained the AI was a “black box”.    For some people, that is true.  It is true for many people who own cars too.  They turn the key and the car starts and they see it as a black box.  If it breaks, they hire a mechanic.      AI can be used that way.  But the guys who built that laundry folding machine are building an AI and see the robot only as a means to verify their algorithms.  Their goal is not a folding machine but an AI that can be very quickly trained to do tasks like folding.  There are no “black boxes,” only boxes where some specific person has not yet looked inside.

All that said, we don’t know if this robot is “real” or if it is being teleoperated by a person wearing VR goggles. 

A commercial product would be VERY hard as it would need to be able to do truly useful jobs on a brand-new home with no programming.  I don’t think this is possible.   I suspect the video was showing either scripted or remotely controlled movements, even if some could have been truly autonomous.

To view this discussion visit https://groups.google.com/d/msgid/rssc-list/60cb30c1-8372-4b09-9590-2f16837f035an%40googlegroups.com.