Simulation of Intelligent Robotic Colony

John Creighton · 2004-01-01 18:36:58

In machine learning the objective is to find a set of parameters so that for a typical set of input patterns, the corresponding output is optimal. If there is one input and one output the objective of the adaptive system may be to find the best fit line between the input patterns and the output patters. The system will learn for a given input around what to expect for the output.

In a robotic colony there are various objectives. In terms of the composition of the population, robots can be assessed upon there ability to locate resources, harvest resources, produce energy, there typical life span etc? On the scale of the robot, the objective could be to decide which rocks are interesting in terms of resources or in terms of scientific curiosity.

Various algorithms could be thought up to decide which robots are the best suited for the next generation and which rocks hold the greatest curiosity. The purpose of the simulation is to test these algorithms and the various robot designs.

P.S. For Inspiration
Rover Pictures

Rxke · 2004-01-01 18:55:35

This is really creepy. I was about to post something *very* similar...

I've been 'doodling' with the fantastic-insect-like BEAMBots, and was just thinking how to do a simulation with bigger 'colonies' of 3-4 differentiated bot-behaviour models... with some kind of central 'hive' system ( ie processing-recharging etc base)

though my progrramming skills are very rusty at best...

John Creighton · 2004-01-02 09:03:30

So I didn't knwo what BEAMbots were so I found an
FAQ:. Are beam boats really smart enough? I think a varitity of robot paradigms should be considered includeing beam bots, rovers, and modular robotics

John Creighton · 2004-01-02 09:26:47

So I was thinking of something slightly more intelligent then beam boats. I know feed forward neural networks are good for pattern classification and I know that human memory is associative. In associative memory ideas are liked together for instance the concept of a fire truck may be linked to truck and truck may be linked to dump truck, fire truck and toe truck. Each unique idea can have its own place in memory.

Attached to this place in memory could be words, sounds, images, and links, to related ideas. Some where within this associate memory there needs to be a pattern classification mechenisim, so that the robot knows which ideas corresponds to which external input (i.e. sound, image, spectrum, etc). The pattern classification can be accomplished by a feed forward neural network.

If the robot can identify simple objects, simple commands can be given to it like, find me blue rocks, don?t go over any cliffs, if upside down flip yourself back up. Maybe BEAMbots will be smart enough. Lets try both concepts.

Rxke · 2004-01-02 09:54:42

If the robot can identify simple objects, simple commands can be given to it like, find me blue rocks, don?t go over any cliffs, if upside down flip yourself back up. Maybe BEAMbots will be smart enough. Lets try both concepts.

This they can, by configuration. (If you build them that way, that is...) They're kinda stupid, but can be amazingly efficient...

(Edit: and lots of low-tech babble deleted... Mr. Creighton's knowledge is far better than mine on this subject. )

The 'modulars' are also impressive, and more programmable to boot.

Rxke · 2004-01-02 10:11:47

Better let Mark Tilden, of BEAMbot fame do the speech:

"Nervous Net (Nv) technology is a non-linear analog control system that solves real time control problems normally quite difficult to handle with digital methods. Nervous nets are to Neural nets the same way peripheral spinal systems are to the brain. This work has concentrated on the development of Nv based robot mechanisms with electronic approximations of biologic autonomic and somatic systems. It has been demonstrated that these systems, when fed back onto themselves rather than through a computer-based pattern generator, can successfully mimic many of the attributes normally attributed to lower biological organisms. Using Nv nets, highly successful legged robot mechanisms have been demonstrated which can negotiate terrains of inordinate difficulty for wheeled or tracked machines. That non-linear systems can provide this degree of control is not so surprising as the part counts for successful Nv designs. A fully adept insect-walker, for example, can be fully controlled and operated with as little as 12 standard transistor elements."

So, Nervous Net=brainless insects. Wich are quite successful (ants, bees,...)
Neural Net: Mammal, birds,.. of course humans. (Wich too, are quite successful)

Rxke · 2004-01-02 10:22:07

As an aside... The two can-or rather should- live together... What we're doing right now with moving robots, IMHO is similar as trying to walk, using *only your conscious thoughts*; 'lift left leg 10 cm., bend forward 5 degrees...' awkward at best. Better use nervous networks for the locomotion, and neural for the real thinking, decisionmaking... commanding nervous nets to 'do their stuff'

John Creighton · 2004-01-02 11:09:51

Maybe, but there are many techniques to control motion. Generalize, but give and work on specific examples.

John Creighton · 2004-01-02 11:27:37

What kinds of robots? Perhaps for production: Arc welders, Hypersonic Cutters, gluers, molders, robots to cover other robots with armor. Perhaps for Science: sptrocpocy, MRI, Humidity, bio detectors, photographers. For agriculture: Weeders, planters, pollinators, watering robots.

John Creighton · 2004-01-02 11:53:12

I think it would be good to start with a factory that can be programed from earth to build a wide variaty of robots. The intial robots could perhaps help gather raw resources. Later robots could be built to aid in production.

Rxke · 2004-01-02 12:05:27

A watering robot would be perfectly possible with a *very* simple BEAMbot approach... cool idea, could save a lot of water, and do a great job of 'micro-managment' in watering, totally unlike sprinklers etc... planting also, if using seeds, not small plants... Weeding should be possible, too, considering each plant has a different IR 'signature' to differentiate the good from the bad... Pollinating should again be fairly simple, at least for 'accessible' flowers (not too deeply hidden underneath foliage)

(Grrreat ideas, John Creighton, you are a great source of inspiration!)

and scientific bots: just add a science package on their backs and let them roam free... Especially in rugged terrain the hexapods are impressive. (they built a simple one, that climbs things that are 2x higher than itself... like stacked books, impressive video... (EDIT... Duh... no video) And i guesstimate it costs oh, 50$?)

About the factory... How big?

Rxke · 2004-01-03 06:45:33

Some picture of the climbing 'bot...Walker V3.3 6 neuron nervous net...
Ok, and these servo motors 'll make it a bit more expensive than my 50$ guess, but it's still remarkably performant for the cost...

John Creighton · 2004-01-05 12:00:27

I?v noticed a lot of papers I seen about beamBots on the net seem speculative and give little information about how these work. They also seem more interested in playing god then trying to find usefull applications. However, beamBots do seem well suited for walking. The design strategy seems to be based on evolutionary algorithims. So I guess to control these beamBots there could be several input that are activated by a remote control. The beamBots that obeys these commands the best will be selected as the parents for future designs.

John Creighton · 2004-01-05 12:05:21

I suppose the first BEAM bots could assemble arrays of solar panels. They could collect them from a factory and carry them away to build beam bot gas stations. Perhaps another beam Bot could be medical in nature. It could go around looking for beam Bots that ran out of power and give them enough energy to make it to the nearest solar array.

Rxke · 2004-01-05 13:09:14

Had to laugh when i read the playing god thing, have thought the same sometimes...

Tilden is an exception, built a 'minesweeper' bot, but ran out of money, the nice thing about it wa that it could take a lot of abuse, like losing legs, but still managed...
Too bad they seem stuck with building 'toys', maybe there are too much complications making them more versatile, or it is too costly to do real research...

Yang Liwei Rocket · 2004-03-11 14:12:05

Yes, why not! Let's pave the way, get our drones to do all that work before man ever puts a foot on mars. This kind of mars project will employ lots of people plus the benefits for aerospace designs and robotics should be great

John Creighton · 2004-08-11 13:05:23

For some reason this is catching my interest again. Maybe it was the DVD I watched on genetic algorithms. Although I new of genetic algorithms before I did realize how effective the are at solving certain AI and control problems. Lets consider how we can cast Beam boat development and an genetic algorithm optimization function.

Thus we need a list of objectives and components for our design space. We may want to start from a particular design. Anyone have a favorite beam bot.

John Creighton · 2004-08-11 13:15:30

Okay, the first objective of the beam boat shall be to get to a location specified by the user. The location will be relative coordinates to the initial beam bots location. The beam boat will have a radio receiver and transmitter. It will receive an AM modulated Signal the message will begin with the word start and with the word stop. When the beam boat processes the message it will return the word acknowledge. The beam boat will then try to reach the destination as quickly as possible. The fitness of the beam boat will be a combination of how quickly the beam boat processes the message, and how quickly the beam boat reaches various distances from the objective location. Fitness will be lost if the beam boat wonders away from the destination after arrival.

John Creighton · 2004-08-11 13:25:54

Some thoughts about hardware. Usually beambots are made by salvaging old electronics. All beam bots need motors. There are motors in many things include VSR and computer fans, razors, blenders, vibrating pens, hard drives, printers, remote control cars. My guess is VCR fans will be the most widely available motor. I wonder what stores sell small motors for a low price.

Rxke · 2004-08-22 12:31:37

I'm currently collecting all 'dead' electronic appliances friends have (well.. the ones with potentially interesting stuff inside...)

Like:
-Cellular phones (the vibrating thingy is a very small motor, and there are some good power-monitoring elements in it
-printers, scanners (motors mainly)
-VCR's again motors
-anything portable, like 'cd-men,' for the low-voltage stuff,
old remotes (LED's) and wireless headphones (IR receivers)
-old disk/CD-ROM drives (sometimes several motors)
-etc... All capacitors are recycled, too, and wheels, springs, etc...

Biggest problem is finding solar-cells: virtually impossible, so you've got to buy them, and they ain't very cheap (the good ones)

Maybe during next year i start building something, thinking 'bout a 'mini-bulldozer' heehee... Got a basic idea scribbled down, but still need some stuff...

John Creighton · 2004-08-22 13:19:19

Biggest problem is finding solar-cells: virtually impossible, so you've got to buy them, and they ain't very cheap (the good ones)

Why not start with batteries. The power source could be replaced later. Besides maybe a highly skilled Beambot could detect when its power is low and then go to a recharge station. The recharge station could be connected, to an extension could, a transmission line, a solar cell, a really big battery, a wind turbine etc.

I saw I design I wanted to try before. It was a spider like walker. Each leg had two motors. One motor lifted the leg up and the other moved the leg forward. I would like though to add remote control aspects to it.

I guess the simplest way to do this would to do this would be to transmit on an AM signal. Demodulate the signal with a capacitor and a diode. Pass the signal through a buffer and then a group of detection filters.

The filter with the largest output would be selected as the operation mode. A Digital to analog converter would then be used to set the voltages that would configure the mode of operation. More advanced designs may interpolate between various modes of operation. Some operation modes may be:
Go forward slow
Go forward fast
Walk in a circle to the right
Run in a circle to the left etc.

Rxke · 2004-08-22 14:29:15

Batteries... Sure. They're being used in the more complex BEAMBots, that or $$$$ NASA solarcells....

But I'd rather have the thingies 'powered' by capacitors, which adds a lot of scope for ultra-quick charging from a variety of sources: batteries, solar-cells, charger-bases...
They're starting to bring fuel-cells for laptops on the market, too... Keeping an eye on that... Probably won't be cheap the first 2,3 years, but might be interesting.

About the RC... Dunno. 'Go left/ right etc... sounds like it would be better to let the robot 'decide' which direction to take in order to go somewhere... Certainly in case of BEAMSBots, they're godd in finding their way, the 'stupid' way, blundering aroud, looking for the least resistance etc...

So instead of go L/R I'd go for 'go to XY coordinates' (and do your thing there...)

John Creighton · 2004-08-22 14:41:23

About the RC... Dunno. 'Go left/ right etc... sounds like it would be better to let the robot 'decide' which direction to take in order to go somewhere... Certainly in case of BEAMSBots, they're godd in finding their way, the 'stupid' way, blundering aroud, looking for the least resistance etc...
So instead of go L/R I'd go for 'go to XY coordinates' (and do your thing there...)

I agree that this is better, but then we need some way to identify the location of the BEAMbot. This could be done by letting the BEAMbot generate a pseudo random signal and then triangulating its position. A more complex system would be to give the BEAMbot GPS and a Kalman filter. Regardless more then one mode of operation/control could be used. Compensations of absolute directive (go to X,Y), relative directives (go five feet forward) and motion directives move forward in a right arching circle could all be used. It is just that the motion directives will be the easiest to give the BEAMbot. BTW even if the BEAMbot is going forward it will still take path of least resistance. For instance it will have no trouble walking over small rocks. But I am thinking forward might not be the best word. How about go north. It would be interesting to see how well BEAMbots coud be made to handle relative directives with very few transistors.

John Creighton · 2004-08-22 14:56:57

The genetic algorithm must design and analyze the behavior of a BEAMbot. The software will first design simple BEAMbots that are just wonders. There wondering behavior and there design complexity will be assessed. Several niches will be established and the beam bots that best fit into those niches will survive until the next generation. After some basic wondering designs are created they will be used as the parents for more complex designs. These more complex designs will be allowed to have things like antennas, digital to analog converters, classification filters, and logic gates. Niches will be established for these more complex designs that include following various types of directives as discussed in my last post. Some BEAMbot designs will be kept because the do one behaviors extremely well. Others will be selected because they can perform a variety of behaviors well. All designs will be penalized for too much complexity.

Grypd · 2004-08-22 15:10:20

I found that to get second hand solar cells the best place is the telecomunications industry. Companies like British Telecom had installed solar cells in its junction boxes back in the early 1990s but have since replaced them with better and added small wind generators. As such there was for a while a glut of solar cell strips.

New Mars Forums

Announcement

#1 2004-01-01 18:36:58

Re: Simulation of Intelligent Robotic Colony

#2 2004-01-01 18:55:35

Re: Simulation of Intelligent Robotic Colony

#3 2004-01-02 09:03:30

Re: Simulation of Intelligent Robotic Colony

#4 2004-01-02 09:26:47

Re: Simulation of Intelligent Robotic Colony

#5 2004-01-02 09:54:42

Re: Simulation of Intelligent Robotic Colony

#6 2004-01-02 10:11:47

Re: Simulation of Intelligent Robotic Colony

#7 2004-01-02 10:22:07

Re: Simulation of Intelligent Robotic Colony

#8 2004-01-02 11:09:51

Re: Simulation of Intelligent Robotic Colony

#9 2004-01-02 11:27:37

Re: Simulation of Intelligent Robotic Colony

#10 2004-01-02 11:53:12

Re: Simulation of Intelligent Robotic Colony

#11 2004-01-02 12:05:27

Re: Simulation of Intelligent Robotic Colony

#12 2004-01-03 06:45:33

Re: Simulation of Intelligent Robotic Colony

#13 2004-01-05 12:00:27

Re: Simulation of Intelligent Robotic Colony

#14 2004-01-05 12:05:21

Re: Simulation of Intelligent Robotic Colony

#15 2004-01-05 13:09:14

Re: Simulation of Intelligent Robotic Colony

#16 2004-03-11 14:12:05

Re: Simulation of Intelligent Robotic Colony

#17 2004-08-11 13:05:23

Re: Simulation of Intelligent Robotic Colony

#18 2004-08-11 13:15:30

Re: Simulation of Intelligent Robotic Colony

#19 2004-08-11 13:25:54

Re: Simulation of Intelligent Robotic Colony

#20 2004-08-22 12:31:37

Re: Simulation of Intelligent Robotic Colony

#21 2004-08-22 13:19:19

Re: Simulation of Intelligent Robotic Colony

#22 2004-08-22 14:29:15

Re: Simulation of Intelligent Robotic Colony

#23 2004-08-22 14:41:23

Re: Simulation of Intelligent Robotic Colony

#24 2004-08-22 14:56:57

Re: Simulation of Intelligent Robotic Colony

#25 2004-08-22 15:10:20

Re: Simulation of Intelligent Robotic Colony

Board footer