Technical background for producing a realistic fembot

[Kube²]

In this fiction I try to give a realistic retrospective point of view of a robotic designer in 2050.

I try to explain what would be a fembot in 40 years.
This is work in progress and I will complete it regularly.
Fell free to react and give your advice
________________________________________________________




Create an autonomous robot means that you need to solve a very small number of very difficult problems. The first one is creating the body and the second one is creating the artificial intelligence.


PART 1
HOW TO CREATE AN ARTIFICIAL REALISTIC BODY
-->SOLUTIONS FOR ENERGY STORAGE
-->SOLUTION FOR ENERGY CONVERSION INTO MOVEMENT
-->MECANICAL SOLUTION SQUELETON AND ARCHITECTURE
-->OTHERS EMBEDDED FUNCTIONS

PART 2
HOW TO CREATE REALISTIC ARTIFICIAL INTELLIGENCE
-->PROCESSOR BASED SOLUTIONS
-->NEURAL NETWORK BASED SOLUTIONS
-->MIXED SYSTEMS BASED SOLUTIONS
-->CONCERNING QUANTUM COMPUTERS

_________________________________________________________

To create the body:
You need to design a system that can move with a lot of liberty during a long time, for this you need to answer 4 mains problems.

Energy efficiency: you need to charge the body or brings to the body a maximum of useable energy.
This energy must be convert inside the body into electrical or mechanical energy by efficient means.
Usually in a robot 90% of internal available space into the body is devoted to this purpose.

Available Space. For integrating all the systems and specially energy storage and conversion systems, you need a maximum amount of available space into the body. So you need to miniaturise every things that can be. You also need to remove everythings that can be remove incuding all the wires by remplacing it by wireless connections.

Minimum weight: if the body is light enough you need to store less energy into the body and this energy is easier to convert into mechanical energy. This is the raison why almost all modern robots are way lighter than a normal human. (For example a 1m70 humanoid robot weight no more than 30 kg)

Overall solidity: The body and specially the squeleton must be strong enough to resist the dynamic effort the robot might support in is normal activities.


The difficulty to find a correct answer to all these parameters is the reason why until very recently a large majority of robots were permanently plug to an external electrical current source or have a very limited energetic autonomy (2 or 3 hours).





-------------------------------------------------
-SOLUTIONS FOR ENERGY STORAGE
-------------------------------------------------

Solution for storing useable energy into the robot

polymers battery such as Li-ion battery:
Such battery reach their limits at the beginning of the 21 century, these battery where used for the first generation of robots a good example is the Honda robot Asimo:
In Asimo (2010) you have around 15 Li-ion battery mounted in parallel. They weight 8 kg (it is more than 25% of the total weight of the robot) they all produce around 50 V and it represent around 1400 W/h. Asimo battery last 45 minutes so the robot need around 1800W/h to function.
It is true that the servo-motors used in Asimo are maybe not the best way to transform electrical energy into mechanical energy. It is also true that some little improvement are still possible but it is very unlikely to see robots based on polymers battery with more than 2 hours of autonomy.

Lithium-oxygen: In a polymer battery energy is store in the form of a chemical couple. These 2 chemical components are the weight of this battery. In lithium-oxygen battery, the second part of the couple is oxygen, which is not a part of the “weight” of the battery because it is “everywhere”. Last but not least oxygen has been choose by evolution to be the energy source of biologic species, it is not a coincidence, it is because oxidation reduction are very energetic reactions. So any chemical couple that include oxygen into the couple has a great potential energy. For instance Li-ion polymers are at maximum 160 W/h/kg and Li-oxygen are in theory
11000 W/h/kg, unfortunately for a number of technical raison (for example oxidation of the air filters) This potential are reduced to “only” 1000 W/h/kg. These battery has been developed between 2005 and 2015. Now in 2050 they are extensively used especially for cars and for robots. For robots they allow autonomy between 6 and 12 hours. In conclusion when you notice sometime that your fembot breath, it is not only “emulation”…

Multi-hydrocarbon fuel cells are available since 2010, but they are here since the beginning. In 1969 the energy in Apollo 11 was provided by a electrochemical cell. For sure in 2050 this technology continue to improve greatly. Now with a full methanol tank your robot can run between 10 and 20 hours. But the “true” advantages of this technology is that contrarily to Lithium-oxygen battery you can run “permanently” your fembot, but there is some weakness in that technology.
A simple fact is that kissing a fembot that just drink is glass of methanol is not so tasty. The second point is that if you run permanently your fembot it can become expensive. Don’t forget the price of fuel when you buy a car, and don’t do the same mistake when you buy you bot.



--------------------------------------------------------------------
-SOLUTION FOR ENERGY CONVERSION INTO MOVEMENT
--------------------------------------------------------------------



Store the energy efficiently into the robot is an important point but not the only one.
You need to convert this electrical energy into mechanical energy.

During a very long time everybody thought that servo-motors were the best way to do that. And that was one of the main reason that explain why robots were so slow and limited in their movement. So, why Scientist were using this obsolete technology ? It is because a servo-motor is very easy to manipulate and program with a micro-controller.
Around 2009 scientist and robotics engineer, understood how stupid they are when they see for the first time the US army tactical operational robot for war zone : “bigdog”.
http://www.youtube.com/watch?v=W1czBcnX1Ww
This day robotics engineers finally understand that pneumatics systems is the only ways to create a reactive and efficient walking robot and things begin to progress faster…

In case you are using a pneumatic/hydrolic systems this mean you need pumps and an hydrolic or pneumatic motor. Such a motor will work with water (hydrilic) or air (pneumatic).
For sure you understand that a water leak into a hydrolic-based fembot is not the best thing that happen !! So pneumatics systems with a air compressed systems are prefered for robots.

The goal of a pneumatic system is to maintain the pressure into the hydrolic circuit to activate with the nominal force all the pump that could be use at the same time into the robot.
Two parameters are important in order to choose an adpated pneumatic motor:
Pound force per square inch in Psi (ie the pressure generated by the motor)
And the size of the whole pneumatic circuit in Gallons.
It is not so easy to evluate that value (I have not acces to the bigdog parameters, and hydrolics systems are not really my area) but I try to make my own evaluation (very approximate) I get :
-betwen 1000 and 2000 psi
-between 2 and 4 gallons
(that might be false so if somebody want to calculate it that's ok)
My point is that:
If my value are correct it is possible to find motor generating 2000psi for 4 gallons with the correct integration level, weight and with a power consumption around 1500 W/h.

This correspond to the electrical power you need to move Asimo but dynamic motion of such a systems a far better than what Asimo can do (see big dog)


TO BE CONTINUED

[D.Olivaw]

As far as energy storage is concerned, you might want to also look into metal-air batteries like those developed by Revolt (http://www.revolttechnology.com/). They are very nice in terms of energy density and power density, both of which would likely be necessary considering the application we are interested in. Another possibility is an array of micro fuel-cells like those developed by MTI Micro (http://www.mtimicrofuelcells.com/).

As to turning that energy into motion, there are also quite a few options. You mentioned servos, of course. The problem with them AFAIK is that it is simple to construct one that will move quickly or with great force, but that designing one to do both (like a bicep or quadriceps stand-in) is very difficult. Another thing you brought up is pneumatic or hydraulic control. These are desirable because of their ability to actuate both rapidly and with great force, but seem to have some issues with fine motor-control. Something I've always wondered about is if this could be mitigated by using many tiny hydraulic/pneumatic "cells" in series and parallel to smooth out the jerkiness inherent in one big one. Maybe someone has more information?

Another thing will be to make sure to budget lots of power for the "brain" of the machine, and to pick one configuration for the CPU and then work out what engineering considerations follow from it. A robot with its CPU distributed in several nodes (with maybe one primary one?) will have a different internal layout than one with one CPU in its head or chest. What method of operation you chose for that CPU is also going to have implications. For a Candy-like fembot, an evolution of modern technology seems sufficient. That isn't the case for more advanced designs. Picking a CPU which is primarily made up of conventional (if cutting edge) electronics but uses molecular-electronics for the heavy lifting would have different results than going with optronics for the processor. We might (for machines with very high capability) jettison the von-Neumann architecture in favor of a true (rather than simulated) neural-net. Whatever you chose is going to require LOTS of cooling; the heat-density in the processor of a modern laptop is already the same as that in the core of a nuclear reactor and a high-functioning android is going to have the same problem only an order of magnitude worse. Optronics could go a long way towards fixing this, as could some sort of liquid cooling system (maybe use the hydraulic fluid?).

[Kube²]

I agree on the many tiny hydrolic motor concept, I see an hydrolic motor based on this concept some time ago (that was for an arm) and it was working quite well.

Concerning CPU I don't think a Von neuman architecture (so something you need to program) will be the solution. There is 2 raison for that
-Because you cannot program every possible situation and every datas the robot have to know.
-Because as you mention it heat density in todays von neumann architectures are already a problem.

Using a hardware neural network is a "natural" solution for all the sensorial needs of a robot (object/sounds/ect) we already know it. For higher function this remains to be demonstrate. But clearly neural network could be an answer to the "cannot program everything" problem, because the sytem could be design to "learn" new problems or situations.

Concerning heat density:
First a neural network is a question of parallelism, contrary to processor wich are sequential. This is the raison why a parallel architecture reach better result than von neuman machine. For exemple a FPGA-DSP (digital signal processing) at "only" 60Mhz is today better for extensive calculation than a i7 at 2 Ghz. The most incredible thing is that you even don't need a fan on the FPGA.
Also, FPGA is a parallel architeture that is a "natural" candidate to implement hardware neural network. (and this is already done, for digital and analog networks ! - neurons could have differents weight not just 0 et 1).
Last but not least the possibility to dynamicaly reconfigure the implementation on a FPGA during execution, mean that you can imagine that if you don't want or can't integrate a very large set of different and specialized neural network in a single chip, you can memorize them into a RAM memory and implement them in less than 2 µs just when you need it to solve a specific type of problem. Then you can switch to an other network for the next type of problem. This is a very new idea but it is an example of interesting approach to avoid massively integrated components, with impossible heat problems...

I also think that brutal calculation power is not the solution for artificial intelligence and that heat density will probably not be a true problem.
In the human brain neurobiologist are understanding that human intelligence is not link to the number of neurons or their density (we are only using a very ittle part of them) but to the way they are organised. The human brain is a patchwork of specialized neuron network (ie for langage, visual analysis, emotions, social analysis ...) and for human beings these specialize system are interfaced in a very specific way with some special retroactions loops that are not existing in the animal world.

The key would probably be in interfacing appropriatly some specialised neural network.
Nature will give us the answer...

[rabiator]

Mechanical drives
For moving the larger parts of the body (limbs as opposed to fingers), advanced servos may still be the best choice. For RC models, there are brushless electric motors with pretty impressive performance available. A random link:
http://www.rctoys.com/rc-products-catal ... ACKER.html
A 500 Watt motor with a weight of only 143 g (5 oz) is pretty amazing, and I'm sure the technology could be used in building some very powerful yet light servos.
For smaller parts like fingers, lots of tiny hydraulics might be better just because of the numbers of servos needed. Building that many motors and gears into your bot might be a maintenance nightmare.

Energy storage
Use whatever can provide lots of electrical energy. Short term that means lithium batteries, long term we'll have to wait what research on fuel cells delivers. I think this is not a problem prospective fembot designers have to tackle on their own, as there are enough other fields that need a good energy source too (laptop computers, electric cars...).

CPU
Here we are probably the farthest from a workable solution, as the research is not even close to produce a convincing AI in a computer of any size.
But while we are speculating, my guess is that the brain hardware of a future fembot will be a massively parallel design, as those seem to provide the best total throughput per power consumption. The whole thing might be one big sealed module where the chips are closely stacked, a bit like in a SSD drive. And it will be prohibitively expensive at first

[Kube²]

Mechanical drives
A 500 Watt motor with a weight of only 143 g (5 oz) is pretty amazing, and I'm sure the technology could be used in building some very powerful yet light servos.

unfortunately no.

These motor are only for scale model aircraft. That mean they turn very fast (big rpm) (for the propeller) and are very light (for the aircraft) but that's all, there is no torque and you cannot stop/accuate them on precise angula. A 30 or 40Kg robot need powerfull motors and power means HP (horse power) HP = torque*rpm/5000...

Now I will show you a "true" servo motor for humanoid bipedal robots:
http://www.maxonmotor.com/media_releases_5807.html
And the application case:
http://www.maxonmotor.ch/ch/en/index.htm creepy Einstein robot...

I have a friend that know the price of this one, I will ask him but It's a less 500$. For HRP-3 (the brother of our "QT" HRP4 ) its 1000$ and there is around 30 servos in it, and they plan to sell it 80 000$.
You see the problem...almost half of his price is just... Servo-motors...

That is an other reason that lead me to think that servo motors is not a good solution comparing to pleumatic systems:

Future of robotics means, affordable, so bankable, so mass produced bots.
I will just take a concrete example.

In a servo motor robot you may have 30 electrical motor. But no more than 3 or 4 are working in the same time
These motor cost 1000$ each because they need to be
-->quite powerful
-->quite fast
-->very accurate (step by step)
-->very small

In a pneumatic system you have only 1 big electrical motor, that create pressure in an hydrolic system, that pressure is used to move the part of the body that "need" it.
These motors could be around 500$ but good ones are around 1000$ too.
-->they need to be very powerfull
-->very fast
-->but they are just turning so you don't need them to be accurate
-->and because for such a pneumatic system there is just one engine you could instal it in the torso, their is no need to miniaturize it.

It is true that good servo motor are very precise and that it is harder to do the same thing with hydrolic, but if you are just waiting for your low cost "candy-like" fembot, you should definitvely bet on hydrolics. Servos are not cost effective!

[Kube²]

I'm a bastard because I've got the bigdog plans

So now I can show why hydrolic systems beat servos motors !



So I don't put all the shematics on the forum but I have the remaining part, So I can describe the system

First, there is a fuel tank
Then there is a motor the important part is that this motor is "just" a simple one cylinger go-kart engine. So you really don't need something powerfull to move the 100 kg of bigdog, indeed the motor rotate at 9000 rpm and mecanical power is only 15hp.
So this motor turn and drive an hydrolic compressor.
With the 15 hp from this motor, the compressor generate a pression of 3000 psi in the hydrolic system.
The hydrolic system is oil. Oil goes into the limb and the acuator of the robot to move it. By doing this the oil gets hotter, so there is an "heat exanger" next to the oil tank to prevent the robot from overheating. :o





beautiful isn't it
So now the question is how could you tranform that into a fembot ?

First you need to remplace a 15hp kart engine by a 15hp electrical motor.
Being the happy owner of an electrical kart I can confirm that such electrical motors exist and that they are smaller lighter and as powerfull than this one . Second you will remplace the fuel tanl by some li-ion battery. Then you can keep the same type of compressor.
And we've got a quiet silent bigdog, wich is not noisy anymore.

I've some more documented stuff specially some values about the weight and the place that are needed for electrical motors hydrolics and compressor, specially in aerospace, I think that might fit.
Work in Progress

[rabiator]

Mechanical drives
A 500 Watt motor with a weight of only 143 g (5 oz) is pretty amazing, and I'm sure the technology could be used in building some very powerful yet light servos.

unfortunately no.

These motor are only for scale model aircraft. That mean they turn very fast (big rpm) (for the propeller) and are very light (for the aircraft) but that's all, there is no torque and you cannot stop/accuate them on precise angula. A 30 or 40Kg robot need powerfull motors and power means HP (horse power) HP = torque*rpm/5000...

Now I will show you a "true" servo motor for humanoid bipedal robots:
http://www.maxonmotor.com/media_releases_5807.html

...

It is true that good servo motor are very precise and that it is harder to do the same thing with hydrolic, but if you are just waiting for your low cost "candy-like" fembot, you should definitvely bet on hydrolics. Servos are not cost effective!

Now that you pointed out Maxon, I found something even more impressive:
http://shop.maxonmotor.com/ishop/articl ... 167131.xml

Unfortunately, Maxon are just as expensive as you wrote:
The above motor costs EUR 682.24 without VAT in the online shop, where the toy from my link costs $65.

Of course, in both cases you'd need gears to trade some speed for torque. The Maxon might need one gear stage less because it starts out slower and with more torque, but it will still need gears.

Measurement of actual angles in the joints of the bot will cost extra, both with the above motors and hydraulics. Neither will tell you by itself the position of the limb.