Knowledge exchange creating walking platform for inmoov

[Connor McLaughlan]

I am also starting to have my try on making a working walking platform for inmoov. True biped (maybe triple legged if all else fails).

What i am interested in is knowledge exchange on parts and methods, as i have seen some people try it already and failed to some extent.

I bring some knowledge from CAD and computer science and had a few lectures in robotics back at university.

What i lack is mechanics and calculating/foreseeing the forces to expect and needed during movement of the body.

My approach of thought and already a little realized construction is to have it 3D printable.

I am trying to use these components:

- 60kg+ servos

- MPU 6050 gyros

- 3D printable frame construction for the legs

- 5 DOF approach (leg rotation, hip, knee, ankle, toes/midfoot (on foot reused from Gaels design))

Any ideas if servos are possible at all? (force, speed, ...) -> I plan to use several of the 60kg+ servos synchronized in a parallel setup

Are the gyros a good choice or is there something better suited than the MPU 6050?

Does anybody have measurements on the reaction time of MRL to servos needed to work in parallel? -> to keep it all synchronized at speed my guess is that it would have to run directly on the arduino where MRL just gives the walking command to be executed).

Just a few thoughts for the fun of it.

[Connor McLaughlan]

I think it is very important to learn from failures. I have seen a few videos of non working inmoov legs and also Gaël's change from a moving design to a static and then a redesign where he tries to make the feet stronger.

Also i see a lot of people trying to use motors from cordless screwdrivers in their designs.

It would be most interesting what approach failed or turned out not feasible and possibly why that is.

After all the question is not if it is possible at all - it is - but if it is possible with a limited budget and involving 3d printing on a hobbyist base...

[Jonathan]

Hey Connor,

At first glance the MAIN problem ist the center of gravity. We humans can change it while walking for example our arms move with us. The point is the servos that are used in the arms / shoulders are kinda slow compared to other walking robots and as of that cann't react in the right speed.

But disregarding these problems I think 3d printing legs is a wrong approach, because plastik is very brittle and weak. I would rather build the Legs out of extrusions similar as James Bruton has done (https://www.youtube.com/watch?v=JWvH5PHKK74).

Although these are big problems to overcome I would find it great to here more about the project!

Cheers,

Jonathan 

[Connor McLaughlan]

Thanks for the tip.

This guy is building really advanced robots. Alsohe opensourced RobotX and OpenDog...crazy stuff.

I will first try around on how far i can get with 3d printing and maybe later adapt some ideas of other projects if one aspect is not performing enough, like speed of actuation, stability of frame or layout of kinematics.

[Andreas]

Hi Conner,

I started with prototyping for walking legs at the end of last year myself and are very interested in some knowledge exchange. You will find my posts in this forum.

 

Regarding the needed forces for an InMoov robot with about 70+Kg to walk, its nearly identical to a human. There is a lot of calculation examples in the Internet for forces on human prothesis. 

I have seen figures, that the Achilles is burdened with 2.45kN while walking and the ankle is loaded with 3.15kN. 

Therefor I stopped designing pure printed parts and stick for "bones". My favorite design principal is the use of  20x20 aluminium extrusion. They that start at the pivot point of a joint and rest 

on the pivot point of the following joint, talking all the weight from the printed hull. Though my InMoov will no longer be completely 3D-printed, but comes still close to Gaels idea of a 3D printed robot.

I started mobilizing the ankles with ball screws, driven by a geared motor. Everything is attached to the aluminium extrusion of the lower leg. After some testing I am not satisfied yet. If the pitch is small, there is a strong force but the speed is very slow.

A short while ago I found the Channel of Paul Gould who build an actuator with a cycloidal gearbox and an brushless motor ( https://www.youtube.com/watch?v=EE5JBO_b6KY ) which looks very fast. 

This type of actutor could be used inline with the pivot point of the joint. This simplifies the construction a lot. 

Alternativly there is the Channel from Skyentific who uses a brushless motor with a planetary gearbox for a robot actuator ( https://www.youtube.com/watch?v=EhJa8kdkzRY ).

Cheers,

--Andreas.

[Vincent Bigiarini]

Hi,

I'm Designing motorized legs from Gael design. Plan is to add some 20x20 extrusion as bone and threaded rod to renforce printed parts.

I have designed a compact compound planetary gearbox and use BLDC motor and encoder. (350kV 300W @12V 1200W @ 24V)

Here is the Gearbox :

And Where I am on leg design :

 

I will use the same Gearbox everywhere except on ankle. Output torque Torque can be 7, 14 or 28m.kg. It depends on pre ratio (direct, belt or pre gear ratio) and motor voltage (12 or 24V) Output speed can be 22, 44 or 88 RPM.

I use CUI AMT 203 V encoder. If I am able to, I will mesure output torque directly in the gearbox body.

[Andreas]

Wow! What an excellent and sophisticated design! Congratulations Vincent!

I would love to hear more from it. 

Cheers,

--Andreas.

[Andreas]

Hi Vincent,

did you build that gearbox yourself or is it available on the market somewher?

--Andreas.

[harland]

I have been working on legs for awhile

video from this week https://youtu.be/l_8YHEU9zG0

my design is down in Autodesk Inventor all 3D

using Arduino mega as the brain for now

some parts alum. and some 3D printed

[Vincent Bigiarini]

Gears are made by EDM wire cut. Gearbox housing is Homemade. So you can’t find it anywhere.

[Connor McLaughlan]

It is great to see a few people working on their designs.

Reading about the required forces, i have some doubt now about my 3d printing approach, but i will try nonetheless until i reach a failure of some kind.

My plan is not come out with a sprinting inmoov on the first try, but with something that is at least moving a little bit and as stable as possible.

Then iterate on the aspects in future versions.

[Connor McLaughlan]

I'm trying to reuse and/or original inmoov parts and keep as close to the design philosophy as possible.

While creating the frame i am now at potentiometer placement. Does anybody know the parameters (and software) to generate the gear from "elboshaftgearV1.stl"?

I need to apply the gear to another part and have difficulties extracting it from the .stl

[Vincent Bigiarini]

What software are you using ?

[harland]

here are some conversions of the file

good luck

[Connor McLaughlan]

Thank you!

I am using FreeCAD and the .stl Converter is not always working properly.

I was able to extract it from the screwable elbow joint: https://www.thingiverse.com/thing:1921536

Also i was trying to replicate it with the gears plugin of FreeCAD, but i could not reach an identical form.

[Connor McLaughlan]

So another challenge with weight has risen:

My prints in PLA were very heavy with 30-60% infill on standard zigzag setting with Cura.

I optimized this to 15 percent infill with cubic subdivision pattern. It has the same strength as before with quite less material. Only obvious drawback is that the infill pattern is clearly visible on the outside of the part. But i'm ok with it.

Another thing regarding weight is that i am using a lot of heavy M8x140 and M10x140 metal bolts. But unfortunately those add too much weight.

Is there anything in between metal bolts and printed bolts, that doesn't weigh too much and is of similar sturdyness as metal bolts? -> I am trying printed bolts now, but i am not too confident that they can withstand the forces of joints in legs.

[Connor McLaughlan]

I found replacements for M8 and M10 metal bolts on thingiverse.

Printing them with 100% infill has not enough strength to replace metal bolts of same size.

So i have to redesign my structures and holes to upgrade from M8 metal bolt to M10 (or higher) printed bolt. Until stability is high enough while having the benefit of reduced weight.

[Ray Edgley]

Hello Connor,

Just started reading the thread and  would throw you some of my discoveries and observations.

First off, no matter what you do, the legs are going to weight almost as much as the rest of the Inmoov robot.
If you scrip on the strength here,it will fail on you.

The robot that is furthest advanced so far within the Inmoov community is Harland's.
He's done a magnificent  job of getting his legs to walk and even turn, albeit with  some failures along the way, but as you said last month, we learn from failures.

The major problem with Harlands design is that it is a static walker, this will limit its walking speed quite a bit.

Having said that, his is also the only one currently walking :-)

I was working with Bartozs for a while on a design, https://www.thingiverse.com/thing:2316843

There were some very good concepts here and also some serious weaknesses as well.

The waist rotator has some weak points and the ankles are also problematic.

In my research, the best thing you can do, is keep the bulk of your mass as high as you can get it.

Work on the assumption, that the legs will each need to support and lift the entire weight of the robot and any payload that it might carry.
The max payload for Inmoov at this stage is around 1.5Kg, unless you upgrade the arms quite a bit.

The best method of movement will be dynamic balancing.

The other problem is the leg must be accelerated at quite a reasonable speed in order to get the foot under the robot before it falls over.

Walking is really just controlled falling. 

This will require a processor a bit more powerful than an Arduino Mega 2560.

You could potentially use something like an  Arduino due, but I think you may be looking for something faster again.

The MPU6050 should be quite suitable for the balance control, however, you  may want to consider where you place it.

In the human body, our balance sensor is in our head,  but  we also back it up with the pressure sense in our feet, our built in inverse kinematics and of course, our vision to determine level.

Ok, so lets learn a bit from nature.

The most powerful muscle groups are in the thighs.

The calf muscles operate the foot and toes on the end of the foot, there are very few muscles if any in the foot.

If you look at some of the faster animals on earth, the emu has all the major muscles way up high in the leg.

So why not mount most of the motors in the thigh and use something like break cables to actuate each of the joints lower in the leg, similar to how Gael operates the fingers from the forearm, but with stronger cables.

Do keep us all informed of your progress, and I do wish you all the best with your endeavors.

Ray

[Connor McLaughlan]

Hi Ray,

thank you for your insights.

Currently my plan is to start small and grow with it and address weaknesses as they appear.

I have not done simulations, but went for double servo setups on each ankle and use leveraging with worm (not direct driving) to see if this can handle the load.

If this fails to handle the weight, then thinking has to start over.

So the kinematics and general structure is finished, i am now at the part where the legs connect to the inmoov low-stomach.

There i am currently designing a construction to do the weight shifting of the upper body to allow one leg to be risen.

If this is a success, i plan to first implement static walking. Assuming there is no obstacle in front of the robot, it will do predesigned static movements, not accounting for any disturbance.

This can be supported by camera, depth and possibly tactile feedback during motion.

If that foundation succeeds, everything else can grow on this base, with respect to it's limits.

The speed of movement will be a problem for real dynamics.

If all fails, i will have at least two static legs for my robot to stand on.

Regards.

[Joel and Katie Hacker]

I believe that the closest model to the human body will ultimately yield the best solution.
Put motors high on the thighs and use connective straps to the lower knee and ankle if possible.
Too much weight down low makes lifting very hard at the hips.

[Joel and Katie Hacker]

I am thinking that an ultimate stability robot would mimic
z the human nervous system where you have accelerometers at
each bone to give numeric feedback locally to each joint actuator
via some sort of feedback loop (like through an arduino)
and then you have mrl talk to the arduino so you have localized
quality control and assurance of stability and localized recovery
in case a foot slips, or the robot is pushed or stumbles

[Joel and Katie Hacker]

Have you tried nylon or aluminum bolts?

[Joel and Katie Hacker]

Well done Ray!

[Joel and Katie Hacker]

Try localized accelerometers as a quality control
to șee if you free getting the desired/requested
acceleration in the right parts, or it might help
give an early indication that something went wrong
and might give the robot more time to recover with
less drastic corrections.... Just an idea....

[Ray Edgley]

Hello Joel and Katie,

What I had in mind for the motor  control was a custom servo driver, controlling a DC motor and using potentiometers at the joint for position feed back.

Think of how Gael modified an existing servo  to place the  feed back pot externally  on the elbow joint.

The custom motor controller is controlled using the I2C bus and provides feedback as to where the joint currently is.

You can see the video of testing the controller here, https://www.youtube.com/watch?v=3vIJzb_Ovzk&t=31s

The controller uses to PID loops to allow the motor to reach full speed and stop reasonable quickly and on target without much if any under or over shoot.
The controllers are based around the Arduino Nano with each Arduino Nano able to run up to two motors each.

I'm not so sure accelerometers is the best approach here, you would need to be performing complex math  real time to see if the joint is moving the way you expected and has a high probability of introducing other errors.
The human body has only two sets of accelerometers, both located in the  head. Every where else is  pressure sensing.

In the Inmoov hand, Gael used 200 Lb fishing line as tendons with all the servos  in the forearm.

As a result,  most Inmoov builders are going to have some (if not a lot) of this line left over, why not use this same line but doubled up to do the same thing it has a 90 Kg load strength two or three of this line contained in a  PTFE tube as a tendon sheath should work very  well.

It is very tempting to build a solid skeletal frame for our robots then cover them to protect the parts inside.
Is this the best approach?
What if out skeletal framework is an exoskeleton structure instead?
It work for the forearm :-)
It provide a solid form or protection and somewhere for the motors to mount while being 3D printable.

Just some food for thought.

[John Stager]

I am doing pretty much what Ray is saying, but buying motor controller boards that allow you to tell the motor where to go by use of encoders on motors. The angles are read by a pot hooked to Arduino Mega. The RoboClaw motor controllers let you set a lot of different parameters including PID, velocity, start and stop acceleration, max amps and a dead zone so you don't have the motor "hunting".  Attached picture is the right foot.  You can not see it in the picture (on the back of the foot) but it also has an IMU on the foot which is being read using an Arduio Nano for roll, pitch, and yaw. I want to a add pressure sensor to the bottom of the foot. 

--
You received this message because you are subscribed to a topic in the Google Groups "InMoov" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/inmoov/WpLSo1BTm9o/unsubscribe.
To unsubscribe from this group and all its topics, send an email to inmoov+un...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/inmoov/3a22ab06-5a9b-48cd-9e85-cbffe92da68f%40googlegroups.com.

[Connor McLaughlan]

Hi Harland and everybody contributing to this thread,

as initially stated, i am trying to build 3D printed legs for the upper inmoov body. I suppose some of you have already tested this and are following a metal based approach now for various reasons. (which could be interesting to hear...)

But i still want to try and see where it might fail doing it the 3D printed way. So i have adapted and used the design principles of the inmoov biceps and applied this to the legs. The basic construction of the first somewhat complete version is done and i have attached pictures at the end of my post of what i am trying to do. So far i have not experienced a game breaking problem, but i always suspect it looms around the next corner :)

I built two legs, one using Hitec HS-805BB and one using JX HV2060MG servos for comparison.

Advantages of servos:

- not too heavy
- torque of two joined servos is high enough for legs to at least raise their own weight
- reasonable price-tag for hobby usage

Disadvantages and possible mitigations for the use of servos:

- slow -> can not react fast enough for dynamic walking. This requires more sensory feedback like depth information and tactile sensors on foot to avoid stepping somewhere where it should not step.
 
- loud -> JX louder than Hitecs due to metal gears. Either find quieter and affordable servos with helical gear or try dampening.

- HS-805BBs tend to move under weight when not powered. Unsure how all reacts to a mounted upper body. Perhaps "parking brakes" are needed when not powered.

Which brings me to the following question:

- has somebody seen faster/higher torque/quieter servos than the ones i am using? (sub 100 EUR)

I have seen servos for about 80 EUR that don't bring much benefit for doubling the price. Then the next stop is between 150-350 EUR for industrial servos which is unreasonable to me and also won't fit the giant scale servo form factor.

The plan is to implement something in between static and dynamic walking. Having defined motion patterns, but adapt them to sensory input like balance, tactile and depth sensor feedback.

Possible Problems i see coming:

- The forces of a moving leg on the hip section are quite severe. I suspect the main problem in handling the forces there.

- The weight of the upper body might be too much for my servo configuration. I have to see how this develops. For the structures i am using Cura's Cubic Subdivision Infill at only 15%, which does wonders to stability and weight of the printed parts. (drawback is that pattern is visible on the outside).

Upcoming steps should be:

- marry legs to upper body
- experiment with weight shifting until the robot is stable and allows for leg raising without falling over
- develop a sensory concept for balancing out when weight is shifting
- develop static walking procedures
- develop a sensory concept for safe leg placement
 

[Connor McLaughlan]

A little update on what i'm doing right now...

I have designed a base where the lower inmoov stomach can rest on without using a pole.

The setup is stable standing without power for now. When the upper weight rises, it might need to be powered or "parking brakes" will be used to prevent servo movement without power and thus collapsing legs.

Currently i am using these servos: DS3225MG everywhere with exception to ankles (JX2060MG) and for torso turning (Dooman DM-S2000MD, because i have them and they are not too slow here).

Each leg can hold and raise its on weight and the whole robot for now.

What i am experimenting on now is a balancing mechanic to be able to raise a leg without falling. The lower stomach should slide over on leg as the balancing base and the upper body needs to work as counterweight for the leg.

[Yohann T.]

Amazing job ! It’s inspiring. Keep us updated, cheers.

--
You received this message because you are subscribed to the Google Groups "InMoov" group.
To unsubscribe from this group and stop receiving emails from it, send an email to inmoov+un...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/inmoov/52a36014-9068-43e6-b86f-9ffea01b0af0%40googlegroups.com.

--

Yohann Taieb , Mobile App Artisan    visit AppsFresh.com

[gael langevin]

Hello,

Very good progress, it's interesting to notice the way you are using the pistons and the bicep parts into your model.

Keep it up!

Gael Langevin

Creator of InMoov

InMoov Robot

@inmoov

To view this discussion on the web visit https://groups.google.com/d/msgid/inmoov/CAGwf9e3K0%2B%3DaMMT4L%3D0tW7RC97SNRfv6rcf%3DggJBMDA5HNpd2g%40mail.gmail.com.

[Ray Edgley]

Just out of curiosity, do you have any  video of the legs moving?

It would be good  to see how  they  move through each of their degrees of freedom.

It would  give us an idea as to how it needs to work so as to keep the balance 

Have you thought about  building  your own servos using geared motors and motor drivers?

This is one i worked on a while back.

https://www.youtube.com/watch?v=3vIJzb_Ovzk

There is a pot on the knee joint, connected to an Arduino Nano.
There is a full bridge motor driver  controlled by the Arduino Nano.

A geared motor with a T8 threaded rod running 12V.

In this case the Arduino Nano was controller via  I2C using an Arduino Mage 2560 as the master (Its it one connected to the computer)

The source code for the Arduino Nano is on Github

https://github.com/Cyber-One/Small_Projects/tree/master/I2C_Dual_Motor_Controller

[gael langevin]

Hello Vincent,

Maybe you had a link that Google didn't like?

Gael Langevin

Creator of InMoov

InMoov Robot

@inmoov

Le lun. 25 nov. 2019 à 22:10, Vincent Bigiarini <bigiarin...@gmail.com> a écrit :

My answer was deleted ???

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

To view this discussion on the web visit https://groups.google.com/d/msgid/inmoov/84567634-d7a5-4fd9-84d6-2381da963233%40googlegroups.com.

[Juha-Pekka Varjonen]

Could we have a 3d printable files for it, thank you!