robo dog, progress or a step back

[Chris Albertson]

I said I'd post progress or lack of same on my dog-bot project.     Some time ago I posted video of a moving 3D-printed leg.   Yes, it moves, but I found I do not like the design.  So  I started on a new concept for the leg.    Several reasons

1) Servos are s-l-o-w.  About 300 degrees per second at best.  There are two ways to make feet go faster (a) make the leg longer or (b) change the gearing.  The shoulder servo pushes a 230 mm long leg as it pushes both the upper and lower parts.  but the knee servo pushes only the short lower leg.  So it is the knee that matters the most.   The belt drive doubles the speed of the knee and raises the weight of the motor above the knee and reduces the rotational moment of the lower leg, a triple-win.

2) it helps if the weight of the robot is not in the legs but above the legs.  So all motors are moved inward by one link.   This means more room is needed in the body so the body is redesigned as well.

3) I want enough range of motion so the dog can right itself when it falls over.  This knee does almost a full 360.

Chris Albertson
Redondo Beach, California

[danie...@aol.com]

Avoid high ratio spur gear boxes that are not back-drivable.

I tried to build a legged system like that 10 years ago. (See point1seconds.com CareIn project)

Belts connected to a high ratio spur gearbox didn't work for me.

The system also needs to be more responsive by collecting and releasing energy.

QDDs driving belts with series elastics make more sense.

Daniel

--
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 on the web visit https://groups.google.com/d/msgid/hbrobotics/CABbxVHsDPnoTO7t%2BH2-4vNe-Ba-1gLsy1vMYjbJaZCqeTB4WXA%40mail.gmail.com.

[Chris Albertson]

On Thu, Jan 6, 2022 at 8:37 AM danielemc2 via HomeBrew Robotics Club <hbrob...@googlegroups.com> wrote:

Avoid high ratio spur gear boxes that are not back-drivable.

I tried to build a legged system like that 10 years ago. (See point1seconds.com CareIn project)

Belts connected to a high ratio spur gearbox didn't work for me.

The system also needs to be more responsive by collecting and releasing energy.

Yes, that Carln biped is unworkable and looks to depend on having huge feet and static balance.   But on that same website, the MJBOTS quadruped robot platform is absolutely the best way to go to make a quadruped robot dog.  However, those MJBOTS QDD100 servos cost $439 each, and you need 12 of them.  So the cost of just the motors is over $5200.  Then you need more than just the motors to build a robot.  There is a lot of CNC'd metal in there.   Buying an MJBOTS robot requires a mid-four-figure budget.

Others sell motors like those for less, but it is hard to get the cost much below $1,000 per leg using currently available parts   I think this is a good research opportunity for hobbyists:  Can you build a back drivable servo using a quadcopter motor and ESC for under $100   But even if you could, a dog-bot would cost over $1,200.  

There do exist examples of dog-bots made with $20 servos.  12 of those costs only $240 and the rest of the parts can be printed

Here is one example of a system like this walking decently.  I read the kinematic software he uses.  It is WRONG.  His dog could be walking a little better but the calculation is not quite right.    He has moved on now but even as-is it does OK-enough

https://youtu.be/S-uzWG9Z-5E

Many of these use primitive kinematic models and are driven by simpler software.  This dog simply cycles the servo joints and makes

no attempt at calculation.   But this "works"  Here is an example

https://youtu.be/w0mLF-rrgPk

Here is one with a better design, notice the lighter legs and his software is better.

https://youtu.be/Sblo40GJvZc

Here is another one, also with light legs, but I did not like the push rods as they give a limited range of motion.

https://youtu.be/ztCE2fqSIxo

I agree 100% that more powerful motors that use much lower gearing are better, better by far.   My plan is to use this $300 platform for proof of concept and software development,  Later if a better motor becomes available for under $100 rebuilding the hardware robot is easy. The hard part is getting it to do some kind of purposeful behavior.

My next-door neighbor's office where he works is buying a Boston Dynamics "Spot".   I plan to go and look at it after it is working.   Their use case is inspecting a large waste treatment plant.

--

[danie...@aol.com]

The BD Spot is about $80,000.00 just to get started.

While Spot is impressive, there are plenty of use cases for an under $10k quadruped.

I never thought people would spend $1000.00 on a phone...but it seems there are millions.

There are already dozens of low cost quadruped kits that don't DO much.

I would love to see an under $1000.00 quadruped with the following criteria :

1. Robust (works for 4 hours a day for one year without major repair)

2. Energy efficient (min. 1 hour run time)

3. Can trot and jump. (shows agility)

4. Carry a payload (do some real work)

5. Has an Nvidia Jetson processor with 4 cameras and audio/speech IO.

I see a lot of hype on the internet for this style robot.

Mostly under simulation, but bots like: https://stanfordstudentrobotics.org/pupper

already are proof of concepts.

The problem is that they don't scale up in size and mass.

This is why we have one-foot tall Robo1 style humanoids for $500.00 but no two foot ones for $1000.00.

But I like the way you think. And if you don't give up you can't fail.

Personally I would like to see a 50mm diameter QDD servo for joints that don't 30NM torques.

That would be a paradigm shift in servo design.

Daniel

-----Original Message-----
From: Chris Albertson <alberts...@gmail.com>
To: hbrob...@googlegroups.com

--
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 on the web visit

https://groups.google.com/d/msgid/hbrobotics/CABbxVHt13g_q%2B36b_4A9e-83tAWPApmt07L%2B8j%2B4v%3D6%2Biv4QSg%40mail.gmail.com

.

[Alan Federman]

while it might be feasible to build a robot like that with a 1k BOM, you wouldn't be able to mass produce and sell it for 1k.  There's development, sourcing, manufacturing, logistics, certification, support, staffing, etc. It would be a major effort to develop a relationship with your vendors who would design or modify the key components, motors, controller, batteries and power circuitry, sensors to source the major components.

To view this discussion on the web visit https://groups.google.com/d/msgid/hbrobotics/1772049586.1192800.1641493348571%40mail.yahoo.com.

[Chris Albertson]

No.   I can't buy the parts even in quantity directly from China to get to $1K for a an MIT Cheetah Clone.   Yes the MJ Robots quadruped is a member of the class of 'bots whose design is based on the MIT Cheetah.   The best price on the is about $3,500 for Chinese vendors who offer typical Chinese-level "support".    The basic parts set price floor above maybe $4k.  I'd like to be proven wrong...

If you could come up with a rough BOM for a robot that is about $1,000.  I'd consider building it AFTER my current robot is walking on its feet.  They all use the same kinematics and computers so the software transfers over.  Actually, that $1k BOM would allow me to build a leg for $250.   I'd do that before the current project is walking on the floor.

To view this discussion on the web visit https://groups.google.com/d/msgid/hbrobotics/1463389124.661452.1641496460986%40connect.xfinity.com.

[Chris Albertson]

The plant where my neighbor works is spending $250K on a Spot robot.   Yes, the robot is "only" $70k but all you get for that is a robot that can follow a remote control.  Much of the $250k goes to consulting work to program the robot to walk on its own around the plant and for specialized sensors to be installed on a backpack unit.   Spot is really the leader now because of its software.   It looks at the gound and steps over things like debris and discontinuities in the floor.  If you command it to run into a post, it will on it's own go around the post.

But there is a market for $6,000 robots.   But I argue there is a use-case for my $300 robot too.   That is software testing and education.

I like you list of criteria.   This is the more useful thing to discus  What can it DO.

I'm gambling a bit that it will work but not gambling much because I'm only building one leg.  It will operate with a weight attached to the foot to simulate a real world load. The rest of the robot gets built only after the test leg works.   Against your criteria I have learned a few thinfs.

1) you have to design to maintenance. the SpotMicro required a complete teardown and rebuild to change a part as the screws for part1 were covered by part2 and so on in a seemingly circular list.   It took 20 minutes to swap a motor.

2) jumping requires lots of money.  It is not technically hard, but the high power density mother cost $$$.  Trotting however is defined as walking using diagonal pairs of legs.   I think that can be done at aut 0.5 meters per second with my loow-power cheaper motors.   Trotting does require supporting the weight on only two legs at a time and BALANCE.   rottiing requires control more than power.

3) specifying a Jetson is wrong (sorry) the thing to do is specify the required behavior and then buy whatever computer is needed.   Jetson is one reasonable solution but so is having two Pi4 and a Corel TPU chip.   The TPU is faster than a Jetsn by a lot but only does 8-bit math wile the Jetsn does floating point.   I thnk I will ensure there is space and power inside its kind of processing   Anoher option is WiFi link to a server.  Boston Dynamics "Atlas" is an impressive humanoid robot but uses a full-on server room for computation.  I am using a 16-core Xeon system with 64GB RAM and Nvidia GPS as the development platform.   (Obviously, it will not fit inside the robot.  But "software first", then choose the processor)

I am thinking of using  the new OAK-D Lite camera.  8 of them would be nice, but the robot has legs and can turn or spin to see more of the world.   The initial behavior I want is "finding things" and remembering what it found and when and where

5) I think today's affordable LiPo batteries will give the performance you ask for,  I think I can get 10,000 milliamp hours in the robot sing batteries design for quadcopter drones.  I am certain a 5,200 mah battery will fit.  But again it is "just money"

6) I am but at all sure about payoad.  I predict 1Kg might work but testing will have to be done.  The servos I use can do 20 kg-cm of torque and the links are on order 10 cm long and there are four legs.  What I don't know is about heat, power usage or how much margin is needed.    But servos with twice the torque are drop-in replacement but at tipple the cost ($60 each)

All that said, SpotMicro is not a bad development platform but for the same cost and the same motors and controller, maybe I can do better.

To view this discussion on the web visit https://groups.google.com/d/msgid/hbrobotics/1772049586.1192800.1641493348571%40mail.yahoo.com.

[danie...@aol.com]

As with so many things, it is asking the right question that is important.

I used the "Jetson" as an example...you are correct, the actual hardware can be anything that can run inference.

My $1000.00 price target was not realistic. For now I still expect the price for a useful quadruped to be around $8k.

My preference is to have an edge device rather than one that needs a wireless high speed connection.

Too many times I have seen WiFi and Bluetooth robots that work in the lab fail at some expo center where the radio space is congested.

I like the integrated cameras such as OAK-D lite.  It makes for less latency...in theory.

As for system energy. Servos are not efficient. The motors are always spinning backwards and forwards during position control.

Artificial muscles are not there yet. But the idea of not using power to hold a position seems to make sense. Again....series elastics help.

Sure you can use a $300.00 quadruped for teaching and experimenting, but these already exist.

What is your goal in "new features" with your design?

To view this discussion on the web visit https://groups.google.com/d/msgid/hbrobotics/CABbxVHsPj7SHj1Q0Z56-RO72gD1S1DkeTF6Dd%2B3Sk0n5NqWccA%40mail.gmail.com.

[Chris Albertson]

I realy do think you could get the cost to user $1,000 but it would take a reach program to find a cheap motor.   Get a 50 mm diameter drone motor that is rated at maybe 600 KV (that is 600 rpm per volt in the intended use case)  These are mass produced in China and the cost is as low as it can ever get.  Then we look at drone motor controllers.  there are design to spiin the motor at high speed to to control torque at low speed but they are just an 32-bit ARM chip and some mosfet's.   I guess that the software in the ARM can be changed and the hardware repurposed.   these controllers or (ESC as they are called cost as little a $12 for one that can handle 20 amps.   Next you need gear reduction of about 10:1 and the only cheap way is to use belts.  But 6mm G2 belt components are now really cheap because of the 3D printer market.   There is a lot of work required.  the hardest is understanding the software in the speed controller and re-writiing it as a torque controller.

Why build anther one?   There s still room for improvement in this space.  Many are poorly designed or the designed have abandoned the project or that t software is simplistic and  not entensable .     The best robots dogs were built by students, not hobbyists but they finish the project, get a grade and move on.

Now to decide on a shoulder.  There are three ways to go (1) just use the servo's built-in bearings.   This kills the servo after only a few hours but is what the first SpotMicro used.  (2) use an above the shoulder brace to hold an outboard bearing,  This is easy and needs only four $1 bearings but adds bulk or (2) use multiple large bearings and a shaft inside a tube.  This requires serious engineering but is what the "big guys"  do, but adds the cost of 8 large 30mm bearings and works best with metal parts, not printed parts.   All three have been done.

After the shoulder then I have to package four shoulders into a body and also allow room for computers and batteries and should plan wire routing, now not later.

To view this discussion on the web visit https://groups.google.com/d/msgid/hbrobotics/1110624228.1464354.1641498482308%40mail.yahoo.com.

[Gmail]

I just bought a dog robot directly from China for about $3000. It runs fairly well. It’s nothing like Spot, but then again, Spot costs many times more. 

Does anyone want to buy it from me?

Thomas

-

Want to learn more about ROBOTS?

https://www.youtube.com/user/thomasfromriverside

On Jan 6, 2022, at 10:22 AM, danielemc2 via HomeBrew Robotics Club <hbrob...@googlegroups.com> wrote:



To view this discussion on the web visit https://groups.google.com/d/msgid/hbrobotics/1772049586.1192800.1641493348571%40mail.yahoo.com.

[Chris Albertson]

It will not be easy to sell at that price.   It appears to be a Spot Micro derivative.

I recognize some details and see how they modified it to use the servo brackets.

It is actually a smart modification.

Could you take more photos with the covers removed to show the structure and

the electronics and is there a link to the software?

If you want to get better performance and don't want to write software from 

scratch, look at this

https://github.com/chvmp/champ

It runs on Linux,  I assume there is space inside for a Pi4.

I should have the above software running in a docker container "soon".

I'm using docker because they require a different version of Ubuntu Linux

and ROS than I use.  But the side benefit of moving to Docker is that anyone

could then run this with zero installation

These servo-based dog-bots use an I2C control chip to run all 12 servos so

there is a really clean interface point, just four wires to connect the

robot to any controller.    So don't worry if they put in the "wrong" microcontroller

it is really easy to change, just one I2C connector.

Again, it would be great if you would show this, covers off or even better

a video.

Someday I will have software.   Today I am making the Python "unittest" package

play with the new quadruped_kinematics package and testing that the package installs with pip. I very

much believe in automated unit testing.

  

To view this discussion on the web visit https://groups.google.com/d/msgid/hbrobotics/76C66BBF-FEFB-45F6-A6B9-5D45340A1E8F%40gmail.com.

[Gmail]

It runs on an Arduino. The seller gave me the source code. Although it’s called a “dog”, the software says it’s a  “cat”. 

Thomas

-

Want to learn more about ROBOTS?

https://www.youtube.com/user/thomasfromriverside

On Jan 6, 2022, at 3:55 PM, Chris Albertson <alberts...@gmail.com> wrote:



<image0.jpeg>

To view this discussion on the web visit https://groups.google.com/d/msgid/hbrobotics/CABbxVHs4_HC50AzyMX7ZjAZ%3D2RcuD2N9LOtw30DwXR_KR1W2-w%40mail.gmail.com.

[Chris Albertson]

Would you happen to have a link to the source code?   I'd like to read it.     If it is open source and no public like is available, I'll make one for you.

As I said, I strongly suspect that changing the processor is very simple.  So which controller board you have should not limit your options.

To view this discussion on the web visit https://groups.google.com/d/msgid/hbrobotics/84838045-8A5F-4F0E-8B10-95CD881E7DAE%40gmail.com.

[Gmail]

There is no link. They sent it to me and deleted the link. 

Thomas

-

Want to learn more about ROBOTS?

https://www.youtube.com/user/thomasfromriverside

On Jan 6, 2022, at 5:16 PM, Chris Albertson <alberts...@gmail.com> wrote:



To view this discussion on the web visit https://groups.google.com/d/msgid/hbrobotics/CABbxVHvbvznzKRHDy39KtkQbrmdP-1VkCr9Ys%2By2TiudLU8O5A%40mail.gmail.com.