Bipedal Robot

[Thomas Messerschmidt]

Hello,

Was anyone here working on a bipedal robot project in the last few months? I seem to remember someone chatting about it. 

Thomas Messerschmidt

[Phillip McCary III]

Thomas:

Sorry, I have not been working on any Robot project lately.

I hope all is well with you and your family!

Best wishes,

Phil

--
You received this message because you are subscribed to the Google Groups "HomeBrew Robotics Club" group.
To unsubscribe from this group and stop receiving emails from it, send an email to hbrobotics+...@googlegroups.com.
To view this discussion visit https://groups.google.com/d/msgid/hbrobotics/CADyjTyaWcawsZZc6hT_n%3DvWg7E2SLJBCBZetmPaDRrsoXybJTg%40mail.gmail.com.

[Dan]

I have been working on a new design.

To view this discussion visit https://groups.google.com/d/msgid/hbrobotics/CAPhOagFb7dNc%2Bp05b_bGyRFCUPYG34iiVrarHr0412TowNVSDA%40mail.gmail.com.

[Chris Albertson]

On Apr 13, 2026, at 10:25 AM, 'Dan' via HomeBrew Robotics Club <hbrob...@googlegroups.com> wrote:

I have been working on a new design.

Maybe you can tell us just a little about it or at least your goals.   

Of course, I was joking when I wrote people are not working on this because they are smart.    It is a hard problem more peopole need to think about it.

[Anthony Andrade]

Hi Thomas, 

I have been developing a bipedal robot and am currently focused on leveraging Isaac Sim/Isaac Lab for policy training.

Anthony N. Andrade

(408)442-9097

--
You received this message because you are subscribed to the Google Groups "HomeBrew Robotics Club" group.
To unsubscribe from this group and stop receiving emails from it, send an email to hbrobotics+...@googlegroups.com.

To view this discussion visit https://groups.google.com/d/msgid/hbrobotics/1E3C5CDD-2440-4C25-882A-48C62203632D%40gmail.com.

[s...@lig.net]

A number of us have that as a goal. We are all generally attacking it from different angles. There are many problems to solve.

I happen to be still mostly focused on innovative mechanical design solutions. I'm driving toward an effective humanoid robot that can be built very inexpensively. That involves all kinds of hard problems. I have some solutions, including a new one just recently.

AI, vision, VSLAM, location, video, end effectors (i.e. hands), etc. are all interesting areas. I also have new hand designs I'm hoping to get to soon.

As far as side projects, I've been part of a streaming VR project / startup-ish group. Three of us have been building that out. I call it 'streaming VR' because it is about using a wearable VR/XR headset to be in a virtual world running remotely, on a beefy server running a high-quality complex environment. I just, finally, finished the new communications layer that provides pretty much the best of any kind of high-performance low-latency link you can think of. High performance even frame + telemetry streaming at high data rates over WiFi with very low latency + tunable FEC, with a nice zero-copy, zero-heap churn C++ messaging library. Runs on Windows, Linux, MacOS, et al. This is just what is needed for an avatar. And many other things.

With that done, perhaps I can dig into the mechanical side. This newest idea is back to my goal of many years: Driving multiple joints with a single powering motor. A while ago I had a method, but it wasn't a great form, so I didn't use it. This idea is a derivative of that, but much easier to implement in a compact way.

Why would you want that? Each powerful actuator + driver is expensive, heavy, and often underutilized. Also, there are a number of cases where you need a lot of independent actuation but don't need strength (facial muscles), or only parts or in total need strength (hand + fingers + wrist). To go more extreme, imagine a robot running on a single motor or engine, like a small gas engine driving all of the joints. There is a standard way to solve this with hydraulics or peneumatics, but those are mostly not interesting for humanoids.

Also, it would be fun to have a wind up robot.

I can't wait to focus on the AI/ML/GenAI side. Of course the longer I wait to dive into that, the further along and easier it will be. So while I keep having ideas on the mechanical design side, I'll have fun with that until I have something capable of useful action.


Stephen

On 4/14/26 8:09 AM, Anthony Andrade <anthonya...@gmail.com> wrote:
> Hi Thomas,
>
> I have been developing a bipedal robot and am currently focused on
> leveraging Isaac Sim/Isaac Lab for policy training.
>

> *Anthony N. Andrade*

> (408)442-9097
>
> On Mon, Apr 13, 2026, 3:49 PM Chris Albertson <alberts...@gmail.com
> <mailto:alberts...@gmail.com>> wrote:
>
>
>
> > On Apr 13, 2026, at 10:25 AM, 'Dan' via HomeBrew Robotics Club

> > <hbrob...@googlegroups.com <mailto:hbrob...@googlegroups.com>>

> > wrote:
> >
> > I have been working on a new design.
>
> Maybe you can tell us just a little about it or at least your goals.
>
> Of course, I was joking when I wrote people are not working on this
> because they are smart.    It is a hard problem more peopole need to
> think about it.
>
>
>
>
> --
> You received this message because you are subscribed to the Google
> Groups "HomeBrew Robotics Club" group.
> To unsubscribe from this group and stop receiving emails from it,
> send an email to hbrobotics+...@googlegroups.com

> <mailto:hbrobotics+...@googlegroups.com>.

> To view this discussion visit https://groups.google.com/d/msgid/
> hbrobotics/1E3C5CDD-2440-4C25-882A-48C62203632D%40gmail.com

> <https://groups.google.com/d/msgid/
> hbrobotics/1E3C5CDD-2440-4C25-882A-48C62203632D%40gmail.com?
> utm_medium=email&utm_source=footer>.

>
> --
> You received this message because you are subscribed to the Google
> Groups "HomeBrew Robotics Club" group.
> To unsubscribe from this group and stop receiving emails from it, send
> an email to hbrobotics+...@googlegroups.com

> <mailto:hbrobotics+...@googlegroups.com>.

> To view this discussion visit https://groups.google.com/d/msgid/
> hbrobotics/

> CADgjOwUx8_uixD9PGW%3DQrDRNdRt01FgvEmXwmztLmfGOQDO3pg%40mail.gmail.com
> <https://groups.google.com/d/msgid/hbrobotics/
> CADgjOwUx8_uixD9PGW%3DQrDRNdRt01FgvEmXwmztLmfGOQDO3pg%40mail.gmail.com?
> utm_medium=email&utm_source=footer>.
>

[Chris Albertson]

> On Apr 15, 2026, at 12:24 PM, sdw via HomeBrew Robotics Club <hbrob...@googlegroups.com> wrote:
>
>
> With that done, perhaps I can dig into the mechanical side. This newest idea is back to my goal of many years: Driving multiple joints with a single powering motor. A while ago I had a method, but it wasn't a great form, so I didn't use it. This idea is a derivative of that, but much easier to implement in a compact way.

Boston Dynamics did that with hydraulics. Atlas could do a backflip. But the cost and maintenance were not sustainable. They went with electric power with 6 to 9 reduction ratios like everyone else. They neded about 6 full-time technicions to maintain about 6 robots. Atlas broke down frequently. Atlas did use an electric pump. Easier robots use gasoline engine to power the pump. But today all-electric is the way to go.

The first-order estimate for a knee joint is this: Some required torque times some required rate of motion. This gives power in watts. Do that, and you might die of sticker shock. So you address this two ways (1) make the robot shorter and therefore much lighter and (2) realize that the maximum rotation rate of the joint does not need to be sustained for more than about one second. The average power is MUCH less than the required peak power. Still, you get a mild case of sticker shock when you see that you need peak-kilowatt-class motors in the legs. “Kilowatt class” is true even if you use steam power or hydraulics. Today in 2026, three-phase electric motors are the most efficient way to create mechanical power. The higher the voltage you run them at, the more efficient. Electric cars run at 400 volts, but newer ones are moving to 800 volts. This reduces “i^2 * r” losses. But 800-volt robots are not realistic. 800V is lethal voltage. A safe upper limit might be 48V. This means a peak of up to about 20 amps per joint. That is very easy with today’s batteries.

The above calculation works no matter what your mechanical design.

The other thing we need is high torque at near zero RPM. It is easy to do this at exactly zero; use a mechanical brake, but leg joints never go to zero if they are actively balancing. Leg joints go from near zero to very fast very quickly but at the tens of millisecond scale. You can’t balance with a slow motor.

I doubt anything today can beat a 3-phase electric motor running at the highest practical volts. There are loads of pretty good robots already working that use this plan. We are seeing convergence as all engineers move closer to optimal designs. The best robots are starting to look alike and for good reason.

The hard part is not the mechanics. It is getting them to do anything useful. Or current AI bubble technology is very poor at planning and learning. To that end, has anyone been following what Yann LeCun is doing? If anyone other than LeCun was saying this, we’d right him off, but LeCun does have a bit of a reputation, and he did happen to invent the technology that is at the foundation of or current AI. His advice to new PhD candidates is “do not work on LLMs, that is not the way forward.” I think if you want a robot to be able to work in novel situations and not require a full-on data center to train, we need to be looking at AIs driven by world models, not LLMs.

I think the mechanics are solved. We’d like it to be cheaper, but mass production will solve that. What would a car cost if you had to hand-make every part yourself? Just making one engine block would mean buying foundry equipment and then machining the casting and so on. But mass production makes cars affordable. Elon Musk is right, someday humanoid robots will cost less than a car, and then like cars, “everyone” can afford to buy one.

I think the target price for a working humanoid robot needs to be about the price of a “good”, reliable used car. If they could do low-skill minimum-wage level work such as loading a laundry machine, digging a hole with a shovel or moving boxes onto trucks, the robot would return its cost to buy in less than 6 months. They would sell as fast as they could be made if the price were about $20K. $20K is Elon’s far-future target price point and for once I have to agree with him.

So I don’t think the world needs $5K humanoids. I think people would rather spend $30K on a better one. Even at the higher price, it is a no-brainer investment. — If it can do useful work. And that is the big problem today.

Look at the car market. You can buy a $20K new car. Toyotoa makes them and they are nice cars. But on average buyer prefer to spend about double that amount for a car. I think this logic will apply to robots.

[s...@lig.net]

BLDC and similar motors are definitely the way to go, especially when they are optimized, have good drivers, etc. And the complexity is not that bad. But somehow even from China, high torque versions are expensive.

For certain classes of robots, all of that is true. That's where most people & companies are focusing. However, there is still plenty of space outside of that envelope. Just in terms of weight, the current approach is not terribly attractive. Also cut cost to a fraction and there will be compelling use cases. Not every situation requires carrying 50 pounds, or carrying that like a body builder.

I look at the car market, and realize that I can buy a really good bike for $1000, an electric bike for $1500, or a functional one for $100. There is some overlap with cars, but they each have big unique zones.

sdw

[Chris Albertson]

> On Apr 15, 2026, at 10:33 PM, s...@lig.net wrote:
>
> BL

> I look at the car market, and realize that I can buy a really good bike for $1000, an electric bike for $1500, or a functional one for $100. There is some overlap with cars, but they each have big unique zones.

I think quadrupeds can do many of the tasks we want. The biggest one is walking on uneven ground that is impossible to drive over with wheels. The design is dramatically easier if you have four legs. This is kind of like using a $100 bike to replace a car.

Those drone motor really can take huge amounts of curent for a very sort time because there is not moving comuttor. Just a solid wire connection. But cooling is the issue. Used in a drine the motoor is under the prop wash and has very high air flow even when hovering

But a robot has a slight advantage over a drone. You can mount the stator to the matal robot structure and use the structre as a heat sink.

Still no mater, even if robots are free, thay need to do usful work. If not everythong else is moot.

[Stephen Williams]

On 4/16/26 12:20 AM, Chris Albertson wrote:

On Apr 15, 2026, at 10:33 PM, s...@lig.net wrote:

BL
I look at the car market, and realize that I can buy a really good bike for $1000, an electric bike for $1500, or a functional one for $100.  There is some overlap with cars, but they each have big unique zones.

I think quadrupeds can do many of the tasks we want.  The biggest one is walking on uneven ground that is impossible to drive over with wheels.  The design is dramatically easier if you have four legs.   This is kind of like using a $100 bike to replace a car.

True.  For a range of tasks, that is fine.

Those drone motor really can take huge amounts of curent for a very sort time because there is not moving comuttor. Just a solid wire connection.  But cooling is the issue.   Used in a drine the motoor is under the prop wash and has very high air flow even when hovering

But a robot has a slight advantage over a drone.  You can mount the stator to the matal robot structure and use the structre as a heat sink.

Yes, and motors could be water cooled like a CPU.  With lots of off the shelf hardware for that.  I've been thinking of ways to accomplish that.

Still no mater, even if robots are free, thay need to do usful work.   If not everythong else is moot.    

Yes.  That will come quickly in several ways.  Especially as avatars.

Stephen