Are humanoids ready for real work? (Agibot G2)

[Alan Timm]

For anyone thinking that the Figure 03 "flip the package over" work unit was a bit simple, contrived, and not realistic (and maybe teleoperated?) check this out.

Agibots working an a pcb assembly line.

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

If this is a ruse it's a very well orchestrated one.

A whole assembly line of pedestalbot version agibots doing what appears to be real work.

These look like AGIBOT G2 models with custom end effectors for the job required.

https://www.agibot.com/products/G2

The future is closer than you think?

[Thomas Messerschmidt]

Hey Alan, have you ever thought of building your own walking humanoid robot from scratch? I’ve been thinking about it myself. I’m sure it’s a long road to get it just walking, never mind the acrobatics, martial arts and dancing.

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 Jun 27, 2026, at 7:23 PM, Alan Timm <gest...@gmail.com> wrote:

For anyone thinking that the Figure 03 "flip the package over" work unit was a bit simple, contrived, and not realistic (and maybe teleoperated?) check this out.

--
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/b14f671c-5b67-4da4-bcb3-f70322bd32c4n%40googlegroups.com.

[Alan Timm]

I haven't put much thought into it.

tbh the hardware is pretty straight forward now, and once you train a good balance/walk/run/crouch policy, walking is more or less a solved problem.

Perturbation recovery is a bit more challenging but from what i can tell they're almost there.

For now Alfie is just as capable as the Unitree R1 and has the same balance problems to boot.  :-)

Alan 

[Thomas Messerschmidt]

I may give it a shot in the next few months. 

To view this discussion visit https://groups.google.com/d/msgid/rssc-list/1948fa6f-1f2d-4c6d-b1bd-729854f44c33n%40googlegroups.com.

[Thomas Messerschmidt]

Do you think it can be done with the larger hobby servos? Or would that be asking too much?

On Sun, Jun 28, 2026 at 3:19 PM Alan Timm <gest...@gmail.com> wrote:

To view this discussion visit https://groups.google.com/d/msgid/rssc-list/1948fa6f-1f2d-4c6d-b1bd-729854f44c33n%40googlegroups.com.

[Alan Timm]

Truth is, you can get anything to "walk" with any type of hobby servos, they've been building small-king humanoids for several decades with standard off the shelf hobby servos.

If you want to do it for real at humanoid size you're going to be using the QDD BLDC motors or (I suspect) that you can do some really interesting things with Serial Bus Servos if you're talking small to medium size robots.  Robotis has some options.  They're expensive but available.

Alan

[Thomas Messerschmidt]

I’m assuming that a taller robot would be easier to balance than a shorter robot because it has more time to react. (fall?)

A few years ago, I built a “life-size “ bipedal walking robot, at least from the waist down. It is not dynamically balanced. It plods along quite ungracefully. There’s a video on YouTube of it taking a few steps.

This time I’m thinking of building something along the lines of Tesla or Figure. Of course this one would need to include dynamic balancing and machine learning. Since the path has already been established, the effort should be something less at this point. 

Still, I assume it’s going to be a large undertaking, not to mention relatively expensive. To keep the price down, I’m thinking of using a number of those 60 kg cm hobby servomotors that I purchased several years ago. Also, I’m looking online and I see some 150 kg cm servomotors as well. No, they’re not serial, but they might be able to work as a proof of concept. (As you probably already know, I was never one to be building tiny robots anyway.)

Thomas Messerschmidt

-  

On Jun 29, 2026, at 10:39 AM, Alan Timm <gest...@gmail.com> wrote:

Truth is, you can get anything to "walk" with any type of hobby servos, they've been building small-king humanoids for several decades with standard off the shelf hobby servos.

To view this discussion visit https://groups.google.com/d/msgid/rssc-list/3257d9eb-3976-4d51-9017-f48374b01e83n%40googlegroups.com.

[Chris Albertson]

On Saturday, June 27, 2026 at 9:02:28 PM UTC-7 Thomas Messerschmidt wrote:

Hey Alan, have you ever thought of building your own walking humanoid robot from scratch? I’ve been thinking about it myself. I’m sure it’s a long road to get it just walking, never mind the acrobatics, martial arts and dancing.

First. Off, if it can walk in the way we can, then dancing and such is also possible almost as a free byproduct.  

You can do this, but there is a high minimum price floor.

Here is the thing why it is expensive:  As a human walks, much of the time one foot is off the ground.  But during this time, we are NOT balanced over the one foot that is in ground contact.    In fact, we almost never balance over it. We fall forward and place the other foot on the ground ahead of us to break the fall and then repeat that cycle.    

If we were balanced (that means with zero horizontal force applied), we would not move.    Isaac Newton explained this: To move our body forward, we need a backward force.  Remaining balanced can’t work.   Making turns and stopping also require that we generate a horizontal force while supporting our weight with only one foot.   Not just horizontally.    We can and do also apply a torque to rotate our body while one foot is supporting our weight.    This is easy; even older and very unfit humans do this.

Here is why it is expensive.   While you can easily compute the average speed of, say, a hip joint.  Let’s say it is only 30 degrees per second.    A cheap serial-servo can do that.    But here is the key— the joint's torque vs time curve has to be VERY FAST if the robot is to keep balance.  Maybe the torque changes direction every 20 milliseconds.    You CERTAINLY need to be able to change the applied torque at about 50Hz control intervals.

Try this experiment:  Stand on one foot with the other only 1/2 inch off the floor.   Your ground-contact leg will be almost shaking maybe 20 corrections per second, but your average joint speed is ZERO.   Nothing moves, but you need to apply horizontal focus very quickly.

So hobby servos are out of the question as they can only be commanded to positions and not to some applied torque, and with 50 Hz PWM they are far too slow to react.

Serial servos are the same way. You send a serial command to, in effect, update their PID constants and even with a fast serial link this is too slow to maintain balance.

Both kinds of servos are controlled by position and you MUCH have control over torque even at zero RPM you still need to control torque and update it about 50 times per second.

EVERY humanoid robot ends up using the same kind of motor, and this is not by coincidence.   They all use 20 to 40-pole three-phase motors where the software has direct control of three H-bridges and also has high-quality rotary encoders on BOTH the motor and the joints. Today, EVERY walking robot uses this.    

Prices for these motors start at about $60 and go up quickly to well over $350.

What many people miss is that a robot controller is working with high-bandwidth torque vs. time.    Typically, a nested torque and position PID loop.   Basically, you need a very fast control loop if the humanoid is not to fall over.     Once you have this, dancing and jumping is easy.