[DemonRaven]

To be totally honest the thought of living forever is not a pleasant one in my mind. Many scifi books have dealt with this subject in detail. Personally i would consider it a form of a living hell. When it is my time to go i wish to rest.

[Ultron]

History is permanent - whether or not somebody will read our books is irrelevant, since each past event has consequences in the future so I suppose it's comforting when I say that nothing you do will ever be pointless... We all leave 'scars' in the fabric of space-time and no matter how small or big they are, they change everything.

And regarding immortality - I believe the sadness of history and past memories would be too much for one man to handle, however I would be willing to go through the hell if it means finishing a life's work and intellectual heritage with the vision of a better future. We might not finish our lifetime projects or live our dreams, but we must leave our children a lesson which will help them achieve those or their dreams and live lives with a higher purpose.

[Art]

I'm sorry but I contend (along with many other more learned physicists and scientists) that non-biological intelligence (AI) will ultimately become dominant.

Humans, should have much, much less than a millennia (perhaps several hundred years) to find themselves obsolete and replaced by a higher form of intelligence (AI). It would be interesting to be an invisible, 3rd party observer to witness how the mech's organize and run everything on a global scale.

I dare say, we shall never know!

No one can say with certainty how long such a mech-society would survive and forward thinking people, at this point, would not care at all!

[Ultron]

It is a high possibility and I agree with your prognosis however I believe that if we are able to create a new form of life then we should be smart enough to be cautious and have prepared a counter-measure - something like that skid windows hack where you set a trap to turn off a PC through a batch-file acting as a Internet Explorer shortcut, but the 'hacker' always adds a second shortcut with an 'abort' batch command in case he/she falls into their own trap.

[Art]

I feel you but I do believe it is already written! Non-biological intelligence will ultimately become dominant! Dominant > verb = dominate = Be larger in number, quantity, power, status or importance. To be in control. To have power to defeat. To be greater in significance than or to look down upon.

Pick one...The future of humanity doesn't look very bright.

The Singularity will happen, computers / AI will become self-ware, then Dominant, then it's over. That's progress. By then, we'll occupy some other globe out in space for several more millennia, if they let us.

The "One way trip" to Mars expedition. Seven years in flight...sleep well my children....BTW, there is no ticket for your return home. (almost sounds like a B-rated Sci-Fi movie doesn't it?)...yeah, me too!

[Ultron]

Before I begin, the Mars One mission is no more than crap. Do mind the offensive word.


Anyhow - yes, the odds are that sooner or later inorganic creatures will dominate this planet and I doubt we will get to any other habitable planets before that.


But this seems as such just because computers are in a constant state of improvement - and as far as I am informed the fields of technology and A.I. are progressing much faster then biology / medicine. But this does not mean that organic lifeforms are not to be upgraded - we have seen what drugs can do to us - from painkillers to 'improved' muscles... And then there are those 'cyborg' projects - we have seen mice been given infrared vision and augmented strength through exoskeletons...


My point is that organic life evolves as well - in a much better fashion. Maybe we are slower, but steadier. Also, with the rise of technology there is also a rise in potential for using it to upgrade ourselves (think cyborgs, but also organic, 'natural' upgrades - genetic manipulation?). No, the human species are not in a dead end - we just can't see through the walls (yet).

[ranch vermin]

If anyone can work out how ordinary childlike logic+willpower works, they are smarter than me.
Let alone the robot taking over the world.  (But I think its more likely terrorists (asian, indian, white, african, christian or islamic, anyone that builds up enough hatred in his guts) could use some dumb slaves to blow things up for them.)


But what im currently into at the moment is "semantic similarity", that I use the last motor that was successful that used the most similar semantic makeup.

What do I mean by semantic -  just a checkbox.  the more checkboxes that match up, the closer you are to it, hook it up to a monkey typewriter system (pull a random motor it hasnt tried yet.)  and it should always keep trying new motors it hasnt before.

So theres a simple mechanism, with an annealing like process for motor development.

However. (hehe)

Setting up these checkboxes to fire at the right times, is going to take a really good inferencing system, that can always spot a man,  spot an apple,  share exact differences in size between things,  make sure at least the classes there are firing.   So even to use this childlike system, its big time difficult, your going to need something that can solve captchas automatically...  just to make its checkboxes more reliable between different states.

And.   The freaking memory system as it gets bigger, the harder it is to make to go realtime,  so youll be trying all sorts of things even trying to get more than 64x64 binary semantics.

Ok, so the thing has NO logic,  but it has semantic similarity instead.  I could easily get more into the gory details,  but ill show you what ive got in a bit.  (im going to use that video tracker, and im setting up a little crawler on my computer- my physics system is totally crap,  i wonder if the brute force process ends up finding a glitch in the motor and hell go shooting off into infinity at lightning speed, because it finds the glitch in the physics system.)

This is where im at, as a nice fundamental platform to go look around for more.

[Korrelan]

Hi Ranch

I have some experience writing genetic algorithm based leg servos.

Basically each motor is driven by a timed feedback loop.  A loop can be triggered into action by either another loop or a sensor.  The main variables required are the time it takes for a loop cycle (servo speed), the duration of the loop and when the loop is triggered.  You have to have a decent physics model so the feet can reliably detect slippage etc, although this can be overcome to degree if joint sensors are used ie. If a servo moves in one direction, the join sensor feedback says the leg has definitely moved but there is no forward motion then scrap that combination. 

Don’t initiate a loop for a motor using a random time variable; select a random position within another loop or a sensor.  Ideally you will end up with a main walking loop, one turn left loop, etc and to walk faster you would just alter the timing of the main walking loop. Running will require different gaits (and will really test your physics engine) but you get the idea.

Also don’t forget evolution, you’ll make life much easier for yourself if you start with short stumpy legs.  Once the neural timings are adequate to achieve the desired motion, extend the legs and let it learn again. Animals didn't pop into existence with log limbs, the evolved along with the neural architecture required to drive them, they swam and crawled first.

You’re also crossing over into reflex actions and DNA based intelligence with this.  How does a new born hippo know to raise its head above water without being taught… simple, the previous generations that didn't have that trait drowned and so didn't reproduce (stretching it a bit I know).

[Ultron]

@Ranch & @Korrelan: Keep going O.O *popcorn crunching*


It is not easy to find references / examples for such specific projects. When I get some extra gear I will set up my own system and I'l hop in to show you what I have been working on.


But this seems easier then it is - I have a load of motors (from phone vibrators to some semi-industrial ones) but very little servos, thus making me either buy H-bridges, DIY Graetz bridges or buy servo motors. With what I have a can *potentially* make a whole robotic arm however details are killing the experiment.


Also having trouble using the cprocessing library for C++ with Code::Blocks, so that's another obstacle...


The investments required for such simple and small-scale projects are ridiculous, even if you have most of the parts. Poor Tesla knew this problem too well (except his projects were not so small-scale).

[Korrelan]

Hobby servos are cheap and easy to convert so they run continually. They have adequate torque for most applications and its wise to standardise, helps developing go faster if you're not having to figure out new hardware combinations every time.

As far as controlling the servos, standardise on one controller. Phidgets make excellent input/ output boards that use USB to talk the host PC at very cheap prices. Phidgets also supply example code i most the popular languages.  Giros, accelerometers, AD/ DA converters, GPS, temperature are all available, I've used these to build the core sensory block in my AGI's head.  I built my first arm when I was 16 many, many years ago. I used stereo tape recorder to control the odd bunch of Motors I'd canabalised from teleprinters etc. Right channel pulse selected the motor and left channel drove the duration/ direction. Oscillator laid the tones on the tape through an old calculator keyboard for programming.  Thankfully things are much easier these days.

[Ultron]

Maybe 1 or 2, or even 5 hobby servos are cheap - but not if you need 17 of them just to start. As my first experiment I would need something similar to the Shadow Robot Arm platform - how cheap do you suppose that can get?


Regarding the controller, I would need to easily interface to a modern computer (so no parallel or old serial ports) and to be able to control it via C++. This is why I have an Arduino (Genuine Uno, but I could buy a $10 Mega from China anytime) - the before mentioned cprocessing library is basically the C++ version of the Processing language which you use to program an Arduino.
These are also very well documented and contain a giant load of examples and tutorials as well as a huge developer community to contact.


Ironically, while my focus is software and I only see hardware as the platform on which I work, the problems I encounter during a project build are mostly in the software. Take for example my Android TV Remote project (look it up on the forum) - I made a functional and handsome gadget which is essentially the 'emitter' of the remote and is plugged into the headphone jack. Why don't I use it now? There is no functional software on Play Store except 1 program - which is limited to 5-6 remotes. They say it can be 'taught' to emulate new remotes through WinLIRC, however this is bogus.


A broad paragraph, but I assume the point was understood. So if anyone has some cprocessing and Code::Blocks experience - I beg of ye! 

[ranch vermin]

Wow, you guys know a bit about this stuff.

Korrelan - have you got a vid of your creature in action?   You sound like you really know what your talking about!

I can sortof understand what you mean by loop, its a repeated action?  but you said 'feedback loop'  and that means something different to me. To me that means putting the output as a part of the input, recycling it.  This thing will detect slippage by not noticing its leg move, but it knows it charged its motor,  so the marking will be there,  but maybe I should have a force feedback value as a vector maybe?
Of course, if there isnt any force feedback, its a scrapped state, generally isnt it.

This thing isnt a genetic algorythm yet,  its more just a little memory system its going to mark off with scores, and head off towards decent scores,  a little like this one (except with legs)->





Ultron - 17 motors expensive?    And even thats not enough!
Its cool you made a remote sensor?   I was thinking making a brain with remote sensors for the cell connections was an awesome idea!  If only the thing would oscillate the messages fast enough.
My little 4 legged creature is all hip joints (sans spin - but if i decide to give it spin, then its 24 motors already, i figured making it all ball joints is more like a real insect, cause they have quite athletic little legs.)
What you cant buy, you have to make yourself, thats what my rule is,  but theres a difference between a little creature in a virtual environment,  then doing it real life,  then the costs come to hit you i guess,  just like you said.




i whipped up a little 3d sim for it,  but he hasnt got his memory yet, but he has his "intelligence" hes kicking around like a fly with a hit of fly spray-> 

Its not really that good to watch, because I havent done it yet, but the physics system was a freaking bitch to write!

https://www.youtube.com/watch?v=BoEwZFJyP8I&feature=youtu.be

* gravity (thats the easy bit)
* surface friction (just done really simply, just sticks him down when his leg isnt going up and its on the surface.)
* ragdoll constraint
* spring length "full hardness" readjustment
* iterating the physics in slow motion 100 times every frame, just to make sure the stupid thing works!

[Korrelan]

I've not taught/ lectured on this subject in 15 odd years, I'll see if I can dig out the syllabus or some code.

Cool video of the bot in the corridor. Very simple reinforcement learning. This works on a very simple principle. The bot has three ultrasonic sensors on the front. Two at approx 30 & -30 degrees from centre and one centre.  The sensors will give a range fairly accurately up to around 500mm through a AD converter. But for simplicity imagine if they return a 0 for no wall and 1 if wall detected.  So 010 = wall straight ahead. 100 = wall at 30 deg left, 110 is wall 15 deg left etc. As well as sensor profiles there are motor profiles. These are usually set randomly.  Around the skirt of the bot are sensors that detect a wall collisions, think of these as pain sensors.  As the bot is running a motor profile incoming sensor profiles are paired as being acceptable unless a bumper/ pain sensor is activated. Ouch... Delete that combination, eventually the bot will avoid walls.  Again uses simple feedback techniques.

My AGI is based on human neural architecture and a similar principle seems to be at work. We don't make smooth movements, think of frames in a video. In utero as we develop our motor cortex throws out random motor signals, we link the servo input with the feedback we get from our joint sensors. Chains of a movement like extending a bicep contacting the tricep are created with the help of inertia, gravity and embryonic fluid. Each frame of a movement only lasts a few milliseconds, and can be altered at any time by sensory feedback. By the time we pop out the servo maps are laid down for 90% of the movements we will ever perform, they just need refining in the real world.  We learn rough servo chains even before we are born, just as the light through the womb wall develops our visual orientation maps etc.  Anyway... Same principle, a servo profile is run and constantly checked against a sensor profile that can change the servo profile, on millisecond time frame.

[Ultron]

Interesting virtual simulations ranch! Regarding the IR sensor brain... Programming and decoding a remote is hard enough, but an array of those things!? Besides IR has some limitations - range, interference (depends on frequency) and it requires an unobstructed view of the sensor (among other limitations). I did not quite understand what your idea was, but I am almost sure there is a more efficient way to achieve that, although using IR modules as 'cells' seems like a really cool experiment, even if it proves to suck in the end


Currently in the phase of theorizing, salvaging parts and learning new languages in order to be able to carry out some experiments without bankrupting. Also, I would DIY my own servos but sadly that would turn out to be the more expensive way (machines + specific parts).


But I am aiming to make the robot head this month - 2 cameras, 2 microphones, 1 speaker and only 1 motor. The brain should be my laptop while the Arduino would simply serve as a replacement for the parallel or serial ports modern computers lack. It will be a bit slow but I doubt I will need quick responses from that servo. Also, the 2 cameras will interface directly with the computer via 2 USB ports (salvaged webcams, anyone?). Will make a thread when finished, but the cool part won't come until I actually give it an arm and a few extra sensors which would enable it to tell what (in the environment) is bad (harmful) to it and what's not, thus enabling it to 'evolve' in a way.


Managed to learn Processing last night so everything is much easier now Gonna have to make my own C++ library for it though lol. Also might have to adapt to C# or brush up on my Java in order to make that TV remote android app and another app to enable the use of an older android phone as an Arduino sensor base - some of you may know that current projects and apps that interface Android with Arduino use the Bluetooth HC-05/6 module and communicate via bluetooth. This however, equals to additional costs and shorter battery life, if the platform is to one day go mobile (doubtfully, but electricity bills ain't paying themselves either).


By the way korrelan, what do you mean by Human Neural Architecture? Can you be more specific, because I have seen a lot of projects based on such an 'architecture' yet they all posses a very different base - this is because in the field of A.I. researches tend to give their own definitions of such a structure. Nonetheless they tend to be interesting and ambitious projects / concepts which I usually enjoy exploring so I hope you do not mind me bugging in

[ranch vermin]

Korrelan->
Thats really cool!    Have you got any vids of your dude walking around in real life?!?  Id love to check it out.   Although I doubt my chances of doing anything past a small sensor pool,  I have a feeling marking it off state at a time is just going to be pointlessly slow.  wise prediction?

Ultron->

Sounds interesting as well.  Id love to see what he looks like.
I guess the motor is his neck right.   Should call him dr no. hehe


Im still stuck on making this memory,  its not so bad making it essentially, cause it could just be some binary backprop thing,  but then I have to go and run it per pixel in parallel and then im stuck,  no magic video games.   But the idea is damn simple, just too computational for my little gtx980 peashooter.