Unitree QMini: An adorable 3d printed biped?

[Alan Timm]

Look a bit familiar?

https://github.com/unitreerobotics/Qmini

https://github.com/vsislab/RoboTamer4Qmini

This guy is AdOrAbLe!  And bears a striking resemblance to another better known Disney robot.

There's only once thing getting in the way of making this one of my next unfinished projects, and it's price.

Those unitree motors go for $369 each right now, and you need at least 10 of them for $3,690.00 total.  That's steep, and I'm not even sure if that's the before-tariff or after-tariff pricing.

https://shop.unitree.com/products/go1-motor

Looking beyond Unitree to other motor manufacturers like Robstride, and you're still looking at $230 each, $2,300.00 total just for the motors (Assuming pricing includes tariffs)

https://www.aliexpress.us/item/3256806644903540.html

I'mma gonna have to wait this out a bit until those motors come down to somewhere more reasonable, $100-$150 tops each.  :-(

Or you can experiment with the full featured simulation and ml training pipeline.  That's free.  :-)

[Chris Albertson]

It will always be the case that bipeds need very high power-to-weight motors.    It is just Physics.      You can reduce the cost as they have done by making the robot very small, but still, you need a lot of expensive motors.  OK, so you have a tiny biped with no arms that can not see a tabletop.   It’s useless except as a cute toy.  It needs to be the size of a petite female human to be of much use.  Maybe 5 foot tall or just under.

Why not build a quadruped?   Those need only 12 motors, servos work, but the performance is very poor.  But for $60 to $80 you get reasonable motors.  The total should come to about $1K.   

I tried making a motor system from repurposed drone motors and 3D-printed belt reduction. It works, but the cost is more than the $80 motors you can buy.  I find that not only must the motors be well over 4X the torque to needed for a static stance.  The motors need to be fast and have almost zero “play”.   

Elon Musk, as we all know, is hopelessly optimistic about future products and even he is saying “Once we ramp up mass production, a humanoid robot should sell for about as much as a car”. SO he is shooting for a $30K robot but admits that is unrealistic today.

Boston Dynamics Atlas is perhaps $1M or even $2M by many estimates.  

<screenshot_05062025_143621.jpg>

<Qmini_simulation.gif>

--
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/6a778df6-0f92-43dc-bd84-19020fc09cdcn%40googlegroups.com.
<Qmini_simulation.gif><screenshot_05062025_143621.jpg>

[Alan Timm]

Something I keep wondering is...  Is it possible to train a policy and reproduce recent results using serial bus servos?

That would put the cost of motors in the $300 range for 10-12 total

https://www.aliexpress.us/item/3256808778910004.html

Scaling down the design for these 30kg 12v servos would be trivial and the resulting desktop biped would be -- even more -- adorable.

Oh, and here's a video of the bot in action:

https://www.youtube.com/watch?v=Q1a-EEGI7XU

[Chris Albertson]

On Jun 7, 2025, at 4:14 PM, Alan Timm <gest...@gmail.com> wrote:

Something I keep wondering is...  Is it possible to train a policy and reproduce recent results using serial bus servos?

No.  It will not work.  The servos are not nearly quick enough.  They are far to slow to balance the biped.  They need to be able to move fast enough to enable dynamic balancing.  This means constant weight shifting to stay upright.

Also they lack the torque.   Remember that a biped has at least 18 motors and drivers and batteries for all of this.   Then to take its first step it lifts one leg and places all the weight of these mototrs and batteries on one foot and then with that one foot the robot pushes itself such that it will fall one stride forward.    But before it falls it places ther other leg in a position to break the fall.  So each leg needs to be about as strong as to hold double the robot's weight.

You want a very high strength-to-weight ratio.  Servos use a small motor with lots of gears.  This gives them high torque but slow speed and high weight.    The better motor, that we need are three phase and at least 24 poles and work with higher voltage, perhaps 24 to 36 volts and have peak currents of up. to around 100 amps.    Yes, 3KW that fits in the palm of your hand.    But 3KW only for some milliseconds. These moters can run efficiently at zero RPM but are typically used with 6:1 planitary gear reduction.  “Everyone” uses repurposed drone motors and “FOC” to control them

One more “must have”. The motor needs to be “back drivable”. In other words, if you bend a joint with your hands the motr will spin.  Servos realy are not reverse drivable.   Finally servos are controlled to a POSITION but what you need to TORQUE control.   You can try and simulate this with a servo mounted into a flexible structure but it works poorly.

This is not a wheeled cart, you MUST do the math BEFORE you start.  You DO NOT want to buy a motor that is too heavy or one that is not big enough.  And you need to calculate the angular velocity of each joint in the worst case and buy a motor with about a 2X performance margin.   Servos don’t have the angular velocity needed for dynamic balance.

We humans stand and walk every day so it seems easy, when walking we push our whole body forward into a planned “face plant” but catch the fall with the other foot.  We keep computing where that next foot should fall until that last millisecond.  It can’t be pre-programmed.

Serial servos do work for a low-performance dog-robot.   Dog-bots have 12 motors and stand on no fewer than two legs at a time, while humans have 28 motors and can stand on one leg.  Humans need stronger legs than dogs.

I’d say build a quadruped.  You will learn just as much as if you built a biped, but at 10X lower cost.   I built a servo-powered dog-bot.   My next robot would be a higher-performance version.  I learned (1) you really have to count grams and (2) every joint needs to be supported by quality ball bearings, do NOT use the motor as a bearing.  (3) You need pressure sensors in the feet. or it is hard to balance or even know if a foot is in contact with the floor

Try and prove me wrong with a simple and cheap experiment:   Put a relly long arm omn a servo, say 8 or 10 inches then place the arm on a table so it suports the arm 8 or 10 inches above the table.  now write some software to self-balance the servo on the end of the arm.  You can place a hing on the bottom of the arm so you only hace to word in 2 dimensions.  When this works, then you know the servo is quick enough and then try and make a stack of 4 servoes self balance.    As it turns out stepper moters are quick enough to self-balance, bnut not serial servoes and not PWM servoes.  Steppers could work bnut they have horrible power to weight ratio.

To view this discussion visit https://groups.google.com/d/msgid/rssc-list/1408cb2d-8a41-48bf-a78c-7eca5f8eadbfn%40googlegroups.com.

[Chris Albertson]

Scaling down the design for these 30kg 12v servos would be trivial and the resulting desktop biped would be -- even more -- adorable.

Oh, A desktop?  Then forgot dynamic stability.  Make the feet very large in the shape of a “C” but facing each other so they interlock.  Then the robot is very stable while standing on one leg, even with the power off.

You only need two servos in each leg.  Four total to make the feet move in a pre-programmed motion.   I have seen simple spring-powered wind-up toy like this.  The trick is to have very large feet so it does not need to actively balance

With such a design, I’’ve seen them use just one simple motoer and gears to drive the feet.  I think I had one of these robots when I was 5 years old.  It would walk on any flat surface.   I’m sure there was no electronics inside, just clever gearing.