My Advanced Realistic Humanoid Robots Project

[Artbyrobot]

I have been chipping away at building some humanoid robots for some years now and wanted to share my build with you all in order to give away my ideas freely as well as pick up some ideas from the community. It's certainly a hard project in some ways. My goal is to have the robots pass for human in appearance and movement. I want them to be able to do chores, manufacture products, and make more robots just like themselves.

Here's the CAD design:

 

[ahsrabrifat]

Looks Gigantic! Is it still in the designing phase?

 

[rjenkinsgb]

Looking at your other posts, you are unfortunately on a dead-end approach, that can never be more than an animatronic type figure.

Artbyrobot - Home of the LarryBot!

www.artbyrobot.com

/diy/ - My Advanced Realistic Humanoid Robots Project - #3 - Do It Yourself - 4chan

My Advanced Realistic Humanoid Robots Project - #3 - "/diy/ - Do It Yourself" is 4chan's imageboard for DIY/do it yourself projects, home improvement, and makers.

boards.4chan.org

That type of setup has also been done already - Inmoov

InMoov - home - InMoov

inmoov.fr

This was it's state ten years ago! He's now looking at realistic skin.

You are working on a common misconception is that human / animal joints can be emulated with small but highly geared motors.

That simply does not work in reality, as biological joints operated by muscles are force-based, not position-based.
The position control is by feedback through your nervous system.

Every joint in a human/animal body can be back-driven by force applied against it, and that capability is essential for lifelike movements and manipulation capability.

Functional joints use torque motors - high power with very low gear ratio that can be back-driven with near zero resistance when not being driven, plus (or torque - current - feedback as well as position feedback.

 

[Artbyrobot]

Looks Gigantic! Is it still in the designing phase?

It should be like 5'9" I think based on my pvc medical skeleton height. Assuming you refer to CAD design, no, I've been prototyping for some time since the CAD was done and occasionally tweak the CAD according to new findings.

Looking at your other posts, you are unfortunately on a dead-end approach, that can never be more than an animatronic type figure.

That's a commonly held view and you may be right. Hopefully, with prototyping and trial and error I can get past any dead ends I am failing to see so far.

That type of setup has also been done already - Inmoov

Yes I'm aware of the Inmoov project. It is really cool. I definitely have a lot of similar principles employed to the approach they use.

He's now looking at realistic skin.

Wow that I did not hear about. I wonder how they can pull that off since the overall silhoutte is not that close to human in their project. They'd have to downscale in alot of areas to pull off a realistic look with realistic skin.

You are working on a common misconception is that human / animal joints can be emulated with small but highly geared motors.

Every joint in a human/animal body can be back-driven by force applied against it, and that capability is essential for lifelike movements and manipulation capability.

Functional joints use torque motors - high power with very low gear ratio that can be back-driven with near zero resistance when not being driven, plus (or torque - current - feedback as well as position feedback.

My highly geared motors I think will be decently back drive-able with fairly low resistance. I'm using pulley based downgearing to achieve this. Although the primary reason was to cut down on gear noise since I want it to do useful work in doors as silently as possible. The gear noise is a deal breaker for me. Highly geared motors with metal gearing are not back drive-able but highly downgeared motors with pulley based downgearing I think are back drive-able but will have to test the backdriveability more to have more insight into this part.

The torque motors you speak of would be nice to use if it were not for the space constraints I am dealing with. I don't think I can use them for that reason in my design.

That simply does not work in reality, as biological joints operated by muscles are force-based, not position-based.
The position control is by feedback through your nervous system.

I'm don't I understand this part of what you said. can you clarify this part more?

Anyways, thanks for this well thought out response!

 

[Artbyrobot]

Here's a finished hand with spandex ligaments adhesive transfer taped and sewn into place. I prefer the skeleton system of a human over the hinge and rod approach although I plan to try to build at least one robot with the latter to compare the two approaches. They each seem to have pros and cons.

 

[jpanhalt]

This appears to be a long term porject that has been discussed on several othe sites, such as:

Other sites:
[URL unfurl="true"]https://www.eevblog.com/forum/projects/my-advanced-realistic-humanoid-robots-project/[/URL]
[URL unfurl="true"]https://forum.allaboutcircuits.com/threads/my-advanced-realistic-humanoid-robots-project.205424/[/URL]
[MEDIA=reddit]robotics/comments/vll03s/my_advanced_realistic_humanoid_robot_project_june[/MEDIA]
[URL unfurl="true"]https://forum.arduino.cc/t/my-advanced-realistic-humanoid-robot-project/1006814[/URL]
[URL unfurl="true"]https://www.robot-forum.com/robotforum/thread/41821-my-advanced-realistic-humanoid-robot-project/[/URL]
[URL unfurl="true"]https://humanoidable.com/threads/my-advanced-realistic-humanoid-robots-project.54/#post-113[/URL]
[URL unfurl="true"]https://www.societyofrobots.com/robotforum/index.php?topic=18209.0#google_vignette[/URL]
[MEDIA=youtube]wNIVozQ3FEQ[/MEDIA]

Progress appears to have been quite slow, but it is a complex project.

 

[Artbyrobot]

This appears to be a long term porject that has been discussed on several othe sites, such as:

Other sites:
[URL unfurl="true"]https://www.eevblog.com/forum/projects/my-advanced-realistic-humanoid-robots-project/[/URL]
[URL unfurl="true"]https://forum.allaboutcircuits.com/threads/my-advanced-realistic-humanoid-robots-project.205424/[/URL]
[MEDIA=reddit]robotics/comments/vll03s/my_advanced_realistic_humanoid_robot_project_june[/MEDIA]
[URL unfurl="true"]https://forum.arduino.cc/t/my-advanced-realistic-humanoid-robot-project/1006814[/URL]
[URL unfurl="true"]https://www.robot-forum.com/robotforum/thread/41821-my-advanced-realistic-humanoid-robot-project/[/URL]
[URL unfurl="true"]https://humanoidable.com/threads/my-advanced-realistic-humanoid-robots-project.54/#post-113[/URL]
[URL unfurl="true"]https://www.societyofrobots.com/robotforum/index.php?topic=18209.0#google_vignette[/URL]
[MEDIA=youtube]wNIVozQ3FEQ[/MEDIA]

Progress appears to have been quite slow, but it is a complex project.

Yeah that is all correct. I enjoy getting fresh eyes on it and giving and receiving fresh ideas regularly. Interacting with communities with overlapping areas of expertise is valuable and can prevent me from getting blind sided by tapping into the collective knowledge of many people which can supplement my limited knowledge and IQ points to hopefully push through the various obstacles.

 

[Artbyrobot]

Every joint in a human/animal body can be back-driven by force applied against it, and that capability is essential for lifelike movements and manipulation capability.

While I think my joints will be backdriveable due to my pulley downgearing, I do have a question on this though: what is the significance of backdriveability for the joints? I'm guessing just so like if it fell the gearing doesn't just break but can turn in wrong direction without shearing gears off? Because if the opposite direction turning is needed, it can be actively reversed in direction to simulate backdriveability even with metal downgearing at high ratios. But if this response was not quick enough in a fall, I guess things break. Is that the main issue or are there others?

 

[rjenkinsgb]

what is the significance of backdriveability for the joints?

"Springiness" - shock loads are a part of normal movements, at anything above ultra low-speed motions.

Also such as balance when picking up or being passed an item, and on uneven surfaces where weight load is proportioned between the legs.

Specifically, both fast and unexpected motions; at anything above a very slow walk, there are inertial loads to be compensated for at every step. joints "spring" slightly with each change of load.

The human/animal muscle "force control" you asked about - eg. If someone passes you something heavy like a bag of cement by dropping it on your outstretched arms, your arms drop to absorb the sudden shock loading, rather than the impact being passed in full. Then your reflexes kick in and apply more force to hold it level to carry.

Likewise when you step or jump off something - you either step on to the ball of your foot and your calf muscles take the impact, or you land with knees bent and your legs bend further to absorb the impact.

Landing from even a couple of feet up with legs straight and flat-footed shocks all your joints badly - much higher and it breaks things! - Muscles are effectively like controlled springs, nothing like normal geared motor servos


The MIT "Cheetah" joint motors are a good example - very high torque with a low ratio (~6:1) planetary gear included, so there is next to no drag when back driving them, unless torque is being applied.

For heavy joints such as major arm & leg joints, if they cannot back-drive just through the weight of the limb and stand impacts of the body weight or sudden significant loads, they will break somewhere in use.
And motions will not be human-like.

A stiff spring connection between the actuator and limb + feedback device can act in a similar way, to some extent for smaller joints. If the actuator itself also has feedback, the difference between the two positions gives a measurement of the force being applied.

ps. If you try to use human bone models, you need to replicate all the supporting ligaments etc. to keep the joints stable in other planes, as well as tendon/muscle equivalents to control wanted motion.
Mechanical joints that can only move in the required axis or axes are far more stable, as well as leaving more room for the control parts.

 

[Artbyrobot]

"Springiness" - shock loads are a part of normal movements, at anything above ultra low-speed motions.

Ah I see what you mean. Like if you threw a bundle of wood into it's waiting arms they have to drop down quickly to absorb the shock load and if they didn't things break instantly. Although if it always expected every load and drove the motors in anticipation in direction of the shock, it could by way of the control system simulate back drive-ability but this would have to be flawlessly executed and that's asking alot but it is possible I think - just would be probably prone to fail at times perhaps.

One thing the Cronos robot did was put inline bungee cords to help handle shock loading, acting as inline springs to protect the gearboxes from those shocks. It's an interesting workaround.

If you try to use human bone models, you need to replicate all the supporting ligaments etc. to keep the joints stable in other planes

True. I intend to do this on robots where I use human bone models.

Mechanical joints that can only move in the required axis or axes are far more stable, as well as leaving more room for the control parts.

True. I will benefit from these pros on robots where I use mechanical joints. I plan to use both methods and compare results.

 

[Artbyrobot]

Here's the ribcage sculpt I did. I am making a fiberglass skeleton for one robot, using a pvc medical skeleton for a robot, and using a rod and hinge hollow steel tubing style design for another robot. I will compare each's pros and cons.

 

[Artbyrobot]

Here's my design for pulley based downgearing. Metal gears I think are too noisy so downgearing by way of pulleys seems best to me.

 

[rjenkinsgb]

?? That is just a pulley block - invented somewhere between 1500BC and 300BC, depending on where you search.

 

[Artbyrobot]

?? That is just a pulley block - invented somewhere between 1500BC and 300BC, depending on where you search.

Correct. However, this design incorporates string as the center axis and plastic discs on the outer sides to hold the line that loops through in place and so it is string based as much as possible with just two plastic discs besides the ball bearing. I've never seen anything like it and most designs are alot more involved and would require a lot more precision machining etc for a housing and rigid central axis rod etc. This is a simple to make design relatively speaking for this small scale pulley.

 

[rjenkinsgb]

The pulley side cheeks will be dragged out of position, down the sides. That's why no one uses that simpler method.

 

[Artbyrobot]

The pulley side cheeks will be dragged out of position, down the sides. That's why no one uses that simpler method.

Yes I can see what you mean. But that has not really been an issue. I've made and tested these but its modified slightly from this design drawing. I'll post photos of working ones I made very similar to this.

 

[Artbyrobot]

Here's an exploded view of a ball bearing and two plastic discs that sandwich it in and the string that will hold it all together for my custom mini pulley design idea.

 

[Artbyrobot]

Side view of a pulley I made.

 

[Artbyrobot]

Here's a double pulley I made:

 

[Artbyrobot]

Here's a turn in place 2.77:1 downgear pulley I setup right next to the motor output shaft.

and here it is in testing: