[HS]

Time to try and implement some of the things we have been theorizing about. Here I will be detailing my first attempt at assembling something that is more than the sum of its parts.

[HS]

General Overview:

I'll be going with the seed principle. Start off small and simple, confirm that it works, then make little improvements, as their necessity becomes apparent from interactions with the environment. The idea is to make the simplest possible version of a whole system, so that it is essentially a ZIP file of nascent abilities which naturally get unpacked, unfolded, filled in, or what have you, by interacting with its environment.

To this end, it seems prudent to adopt as a starting point, the rough replica of an early life form. A small organism ambling its way through the early oceans, generally leading a simple sort of life. You know, stopping to smell the sunlight when it found some, and vaguely knowing to avoid the dark abyss if it could help it.

This idea will probably take the form a box with a few starter systems in it (brain, batteries, motors, electric valves, etc.). This box will move itself around on legs. I feel that wheels are too easy and would stifle development. It would know where to go from light sensing antennae. It would know where it can not go from touch sensing antennae.  On top of the box it would have several solar panels with adjustable tilt.

The creature would essentially be a physical neural net tethered to the environment through a few senses. At first, it will be a basic nervous system, making the creature an ABI or Artificial Body Intelligence. If things go well, this will develop in complexity.

[HS]

Preliminary Specifics:

Body: See if you can make heads or tails of this drawing.



Nervous System:



Composed of strings of nodes strung out like washing lines. The nodes are assembled from:

1. Arduino programable microcontrollers (8 or 4 pins, not sure yet)
2. Tiny Solar Cells (shown below)
3. LED's (shown below) of three colors (red = communication, blue & yellow = emotional feedback as a basic guide to actions)






Muscles:




What else?

Batteries: 9-volt rechargeable for the brain, and small RC Li-Po for motors.
Motors: Small strong brushless for RC stuff.
Structural: Wood, bamboo, basic door/chest hinges.

[LOCKSUIT]

Is it a RL body robot, or a net on a chip in a sim ?

But the brain part is important first....don't go for a body yet....

It has to learn sensory, like see or read about its room, and then predict what may happen and discover solutions using old ideas it can use.

It needs a goal-belief system, that works on an agenda.....not learn a muscle set that naviagates.....its need a language, for its agenda

[Art]

Interesting. What is your reliable method of connecting all those tiny solar units together and what is the output rating of each solar square?

[HS]

Is it a RL body robot, or a net on a chip in a sim ?

RL of course! It's the way to go IMO.

But the brain part is important first....don't go for a body yet....

Nope! I believe a body informs brain structure and development.

It has to learn sensory, like see or read about its room, and then predict what may happen and discover solutions using old ideas it can use.

I trying to start from the very basics. Mine will learn from random doings to make doings less random. I don't want to build a working system, I want to build a system that will learn to work by its self.

It needs a goal-belief system, that works on an agenda.....not learn a muscle set that naviagates.....its need a language, for its agenda

It will have a basic pre-programmed emotional agenda. It should create a "rational" agenda by its self. Such a simple organism won't be able to do anything with a complex language just yet, that should come later.

[HS]

Interesting. What is your reliable method of connecting all those tiny solar units together and what is the output rating of each solar square?

Lots of delicate soldering. ~250mV and 47μA, I might need some op-amps.

[WriterOfMinds]

Independent, multi-joint legs are hard.  I don't know if this is your first robot build, but if it is, I would strongly recommend going with wheels or another type of simple drive train (whegs, tracks, a crawler drive that runs multiple rotary legs off a single motor).  That way you can figure out motor control, which can be enough of a problem on its own, without worrying about joint design, balance, mechanical advantage of the muscles, numerous degrees of freedom, blah blah blah.

I've worked on a quadruped before.  If the frame is not rigid enough or there is too much play in the joints, you might not even get it to lift one foot off the ground without tipping over.  Just so you know what you're getting into.

Have you picked out (or designed) your motor controllers yet?  Have you thought about gearboxes or other ways of trading speed for torque?   I notice you did not mention any position or tension sensors associated with the leg muscles and joints, or touchdown sensors on the feet ... these are a must in my opinion.  The brain will have a tough time guiding the legs in a useful way without sensory feedback on the state of each segment.

How many neurons do you think you'll need in the brain, to learn and control a quadruped's walk cycle?  How many fit in your time/money/space budget?

[HS]

That's really useful feedback. I don't know about most of that yet. I'm still gonna go with legs though, it's a hobby so I don't have a deadline or budget. Who knows about my future finances right? Basically, it's OK if it's a big disaster.

[HS]

Figuring out the brain:

Basic functions such as autonomic body operation, and curiosity, manifesting as explorative behavior, will be hard-coded as a mental foundation. The global thought pattern will be a narrative algorithm. Goal, conflict, result, emotion, reason, anticipation, choice, next goal, and so on.

The goal will arise from the needs of the body and solutions in the environment. Conflict arises naturally as it is unfortunately the prerogative of obstacles to stand nonchalantly in front of solutions. Results can be a worsening of the problem, a situational non sequitur, or an improvement. Emotion lets you know if the problem intensity trend-line was bad, neutral, or good. Reason stands on the shoulders of emotion, gets pointed in a productive direction, and is set loose on the facts of the situation. (The emotion provides a nice vague boundary to prevent the reason from coming down with intellectual over-extension, or the cure for it, self-referentialism.) Anticipation climbs on to the shoulders of reason to see even further. It estimates future changes to the situation. The most important events are anticipated the most. Importance is a measure of predicted positive/negative effect. Importance dims with temporal distance. The next goal seeks the brightest solution, and so on.

Each main stop on the algorithm will probably require a bunch of sub algorithms.

The artificial neuron ideas are also progressing. At the moment it's rubber cubes. The contacts (slightly raised) are magnets, four on each face, five on the power sides (opposite each other). The flexibility of the rubber will allow all the magnets to make contact. The spaces between the cubes will be used for cooling, if I find a non-conductive non-corrosive fluid for both iron and rubber. Such a structure would be simple to understand, easy to power, forgiving of impacts, maintainable, and compact. On the inside, microcontrollers, small batteries (redundunduncy), wires, air.

[goaty]

using legs instead of wheels would test your artificial mind more.

[HS]

Like This:



Held together with magnets, (far apart magnets are parallel power supply, regular spacing = input/output)

[HS]

Does anyone know how to solder a tiny magnet without demagnetization? Also, does anyone know of a mechanically flexible substitute for a thin wire, which could be moved around without breaking due to metal fatigue?

[Korrelan]

I can see several problems with the above cube design, not least of which you won't encounter until much later.

The pins have to be sprung, fitting one or two in a line will be fine but as soon as you come to fit four or more in a block, unless the sides and pins are 90 degree accurate to within a least a thousandth of an mm they won't make contact.

 

[goaty]

Thats it, this excellent work is sending me rambling on now!!

welcome to the long sad journey of reinventing the wheel.   not I, but I think If anyone solved artificial intelligence they are not telling anyone they know. 
That's what a.i.'s about,    I believe in you! and us!

IMO I think you should not bother with making the body react so perfect,  the mind is more important,  if it can get over little quirks in its design its a more adaptable machine.   Maybe packing it with sensors at the end just to be over safe is a good idea,   but really the least sensors you NEED to use the SAFER the machine is!

Don't let the failures of the 70's to generate powerful ai let you down,   the old idiot "if it hasn't been done before that's the reason it cant be done"  is a load of utter toss, you may as well start thinking arrogantly... I think us younger generation should know we outclass the prior generation in all fields,  its not the field that's bad,  its was the idiots running it.    Maybe family men (our parents) aren't actually the same thing as really passionate inventors.    Why prepare life for your child when you can perfect it in your own lifetime?

A replacement for wire that is stronger?  For its skinniness I think it would be quite one of the strongest things.   Maybe ribbons would be stronger but maybe it doesn't suit your design.

Soldering magnets sounds fiddly to me,   im not sure its a good idea but your soldering skill ability will go up a notch.   my best advice there is temperature and how long u leave it beaded, and good timing get nice goobs to stick on.    Solder isnt the toughest metal, you can actually bend it with your hands once it gets grabbable enough.   Just make sure u get nice wiffs of the fumes up your nose as you go, its like napalm in the morning.