Aiko hardware and software BRAINS

Hi

Aiko's hardware:
It's hard to say, because I always change the parts inside of Aiko.
Aiko will undergoes major improvements over the next 2 months.
2 Mainboards, 8 micro controllers, lots of sensors, new servo, new actuators, cameras, mic, new voice engine and much much more.


Aiko's heart and soul is controlled by the BRAINS (Biometric Robot Artificial Intelligence Neural System) software.
It took me three days, before I figure out what to call it (the old name was Sandai).

It's written in C# and Basic with many endless nights without sleep. It took me years to get this far, and there is still more work.
For more information:
http://www.projectaiko.com/technology.html

Many of you ask, if the BRAINS software ever going to release as open source.

The answer: Maybe in the future.
I never do any documentation when I program, out of 800,000+ lines, I think I have about 6 lines of documentation.
Not too mention, I have a lot of code to clean up, if I release it as open source.
Right now, I don't have the time.

The only other robot beside Aiko to use the BRAINS software is Intel's Robot ARTI.
ARTI runs solely on the BRAINS Lite software.
It's used simple version of the BRAINS software (1/10) technology.
I had fun programming ARTI.

You can see ARTI in Intel's Museum in CA. USA.

Watch Arti here:
https://www.youtube.com/watch?v=1jv_aIWZPts

First to introduse myself: Yordan Iliev from Bulgaria. I have many hobbies, but I don't put much effort in each one. Recently I returned to coding chars for M.U.G.E.N. (in case you don't know what M.U.G.E.N. is click here)
Up until now I didn't post anything, because I don't know anything about robots and as for anime - I'm more of a "Naruto" fan. So I've been wondering with this software if such thing is possible: To have a % chance that the robot in question will do what you order it. And to increase the % if you say "please"?
That's it.

Hi ShinzuiUindo

I am not sure what you want.

"And to increase the % if you say "please"?"

Can you explain to me again.
Thanks.

Sorry, English isn't my native language.
Okay, for example if you say "Aiko, feed me" (like in the video) Aiko could deny and say, like, "feed yourself" (around 15% that Aiko will deny?), but if you add "please" ("Aiko, feed me please"), it will be less likely that Aiko will deny.
But now I think that this will not be possible or verry timetaking, since the only way to do this might be to set "......., please" as another command and there would be twice as much commands. But still, I'm not familiar with robots. I might be wrong.

Hi

If you say "feed me" Aiko will feed you 100%.
However, if you keep moving your head like a kid who refuse to eat, then Aiko will know that and will say
"feed yourself".

No, no, no, I mean: Is such thing possible?

In theory, no not possible with the current technology.
In the future when Nano Technology... it better then maybe/.

Dude, you are like, "The God Programmer"....
I'm a software programmer myself, but I can't even imagine how you coded just 1 of all of those components that Aiko is built with....BRAINS seems insanely complex.

This may be a tremendous border breakthrough between human and robot. I'm very interested in this development. Unfortunately, I can't quite contribute anything to this just yet...but will check in from time to time to see how things are going for you.


I wish good luck and success to you and Aiko.



X

I really impressed with your software.Because of this,I decided to build part of your software for my graduation project.Sry,I forgot to introduce myself.I am mohamed,spending my last year in a faculty of computer science.So,could one year be enough for building part of your software? and could you help me if i stuck in the middle of my graduation project?.
Thanks.

What is the best camera available right now for using it in my grad project?,but also not cost me more .
thanks.

Hi hazaza,

I suggest you use Sony camcorder with firewire or DV out.
You can get here in Canada for $350 at best buy.

I didn't know you were the one who programmed Arti, and made BRAINS. (which unbelievably awesome)
If i wanted to create a AI system is there any base GUI code i could use similar to your BRAINS.
Or do I have to make it completely my self from the ground up....
or is there perhaps an old version of brains available anywhere? u are the master programmer!! plz lead me in the right direction.

Hi

I don't think there is any robot software out there.
The only one I know is Microsoft Robot Studio..... I have never tried...
You might want to check that with Microsoft.

What speech engine and voice are you using for Aiko? I've never heard anything that realistic before.

Hi,

Nothing special for the speech.
SAPI 5, with Lexicon.... you just have to do the voice engine yourself.....
The original sound you hear was done in Basic.
I think Microsoft Robot Studio comes with voice engine....

By way of introduction, I'm a DJ, NOT a computer person! I'm also a part-time writer and am working on a science-fiction novel that features robots of all types.

That said, I'm looking at your chart of the BRAINS software and noticed there is a "learning" unit. Can Aiko / BRAINS actually learn from its experiences? Can it self-program to an extent as humans can? (Or is this a goal for future incarnations?) To me, heuristics is one of the fundamental differences between human brain function and computers. Any kind of true AI will need to have such a component IMO.

Please forgive my ignorance if you've already addressed this, or if I'm somehow off the mark. Just trying to get a grasp of where you're headed. Your project is very inspirational! Thanks!!

Aiko does have basic learning, not advance as human....

There is no such thing as a true AI... it's impossible to do that now..... it will never happen...
Making a true AI such as a Human is like building a Time Machine, it will never exist or made.

Hi there.

Is the BRAINS software capable of interfacing with any kind of recognition software for the voice and vision or is it a standard fixed interface thing. Reading that C# is used thus that means your using windows operating system or linux with the mono platform. Either way I like the brain system you've single handled solved a problem that I myself and I think most of us struggled with and thats the way Brains handles the soft and hardware devices.

I also know that people think your a nutcase if you say to build something of this scale.
To claim that your working on an android which is a multi component system with some advanced form of a.i and prove it is very hard although you somehow manage to have a good start so my applause for that.
I am working on my own a.i system for a couple of years and have still a long way to go but still missing a lot to prove my own claim for the sake of skunkword I called it wulfcry Idoru system a work in progress. This a.i system is ment to understand any aspect of the process its collaborate with a user. I've blog in a messy note kind of writing a lot of my view and insight of how some a.i processes should be shaped.

And am very suprised how brains handles the component it communicates with. I've never solved that part. The main purpouse of idoru was continuous recognition. So that you could form idoru to your own taste of intelligent behaviour.

I've also read comments about which hardware to use or that could be build my own idea where not to invent the wheel again by using technology that already proved to work and only needs to be interfaced like canon photocamera have recognition software which u can program its called CHDK thats one of the many solutions.

I wil also let anyone hear about AIKO through my blog. (rejects rule the world).

Hello,

I would like to know how well the BRAINS software would work in a robot like the WABIAN-2R

http://www.takanishi.mech.waseda.ac.jp/research/wabian/index.htm

This more like the kind of robot that I have been trying to build for some time now.

Asbrandsson

to: WulfCry

An example of Aiko simple learning abilities are:
Learning new words... If any thing she doesn't understand she will spell it out like a 5 years old kid asking for help.
Then I will teach her how to pronouce the new word in her database.

This is one example of Aiko learning abilities...



To: asbrandsson

In theory the BRAINs software will work with all robots.
Just need to modified the code to work with their micro controllers.


I hope that answers both of your questions..

I also was inspired by the CHOBITS anime back in the summer of 2005. My ambition focused on the conversational agent issue. Popularlly known in english countries as chatbots. For those of you already working with Anime or CG, rapid development in the past 2 years had created a number of open source 'AVATAR' interfaces designed to interface with the 'ALICE' series of open source AI engines origionally designed by Dr Wallace of England. I believe a FLASH avatar is available now also. If you set up a modular software design, you can create a avatar IO that can later be applied to hardware. A good example of the state of the art is a propritary product by Haptec Inc. Hardware costs lots of money. Software design (assuming you have the tools) costs nothing but time. And I do mean a LOT of time. Creating your character requires about the same effort as writing a novel. Chatbots rated as superior by peers usually have over 12,000 phrases they can say. Their is also a number of open source "Speech to Text" (STT) and "Text to Speech" engines that you can use. (voice track audio) For most of these freebees the STT works poorly to moderately well. If you have a commercial application in mind I would strongly recomend Dragon Naturally Speaking Ver 10. And NOT any of the older versions. Note the developer's kit is over $1000 US. Getting a leg up on the phrases issue (things for the bot to say) The ALICE website offers the "superbot" data base, this is a spread sheet of the 10,000 things most people say to a bot. I think the price is less than $100, fully editable of course. Now in the virtual world of "Second Life" you can find a number of examples where 'botmasters' have put all these elements together into a anime avatar chatbot. Limits in Second life do not currently allow facial expressions. However, Sony Inc's new virtual world does have facial expression controlls for it's avatars. Most all the propritary avatar software tools have elaborate facial expression control features. Now after saying all this, I have to admit I myself do not use the Alice AI engine for my chatbot Chichan. I chose a propritary product that has fundamental design differences from Alice. I will explain in a future post. The news is all good. AI chat engines appear to be progressing much faster now than in the previous 20 years. Software tools for chatbots is also expanding rapidly now. And there is even two software languages under development specifically for modeling instincts and emotions.

Thanks for the answer thats quite an advance learning ability it reminds me of NETtalk a similar kind of speech recognition program created at John Hopkins University somewhere in 1989. I never knew that such an advanced progressive training method would be used because its need a lot of processing power adjusting the neural network from talking babbels into correct pronouncement.

Currently the most successful adaptive learning AIs' are combinations of neural network algorithms and statistical modeling databases. This can be seen with last year’s winner of the DARPA road challenge. Now most of the other top 5 autos also had statistical pattern matching algorithms, but the winner had an additional algorithm that acted as a noise canceller. It was able to recognize noise from valid data and block that from downstream processing; greatly improving real time performance.
Another example of statistical AI is Google's Chinese language translator. It actually has very few grammatical rules programmed into it. It has studied thousands of professional translations and built a database of phrase matches. Google's Chinese translator is rated 98% accurate; better than all the 'hard coded' translators before it. Dynamic learning behavior can now be modeled with a hi degree of reliability for real world applications. The trouble is these leading edge AI learning engines are corporate jewels’ held under maximum secrecy. Tiny pieces of the math are public domain in research papers; but you would have to be an extremely good researcher and statistician to bring the pieces together. Only one of the AI chatbot engines that I know of is based on statistical modeling learning algorithms. That is Rollo Carpenter’s Jabberwacky. It has won the Lobner prize twice in the past five years. Compared to the other chatbots it has some strong advantages, and a few disadvantages. Differences in the performance of the AI chatbot engines I will discuss another day.

Its very interesting to discus this subject but I rather build and use what is available to construct,
A system that is near A.i. Looking into the local bookstore there is tons written about automated learning,
robotics, neural network theory's and the math behind. The net has millions of links to research of the lather.
Two years from now we could still discus which development has tricked out some feature as to robotic advancement.
In the mean while development takes its course to improve computing structures.

I want to know how well could those lets me call them ''components '' be used to assemble a awesome system that simplify,
A.i to Bragg about.

If it where for me I want to be the next Steve Jobs saying no thats not how a computer should be lets build it like this and that
put these nice features in it and TADAAA a Macintosh. It can be done with robotics as for A.i system software.
Necessarily to go as i've describe it here does not have to be. But for what I have and know of and working on wanting to beat,
that Status Quo of robotics build and software .

So my point of view is if some of YOU can find so much that would benefit A.I, if YOU have a "vision", if you think that is going the right way stop talking stop looking stop finding stop reading? and GO GO GOOOO.
build something to Bragg about it I'm going to do just that.

The only thing I need now is a mini AIKO to work on so Tre keep one in storage for me.

Aiko wrote:Hi

Aiko's hardware:
It's hard to say, because I always change the parts inside of Aiko.
Aiko will undergoes major improvements over the next 2 months.
2 Mainboards, 8 micro controllers, lots of sensors, new servo, new actuators, cameras, mic, new voice engine and much much more.


Aiko's heart and soul is controlled by the BRAINS (Biometric Robot Artificial Intelligence Neural System) software.
It took me three days, before I figure out what to call it (the old name was Sandai).

It's written in C# and Basic with many endless nights without sleep. It took me years to get this far, and there is still more work.
For more information:

I was thinking about this the other day, about how a robot perceives the human world.
First I want to talk about learning as it's depicted in your diagram.
In the Youtube video, you told Aiko where the washroom was and asked her to repeat it.
This is thought how an infant learns, but it actually learns by environment, not repetition.

For example how a baby can tell the difference between his mother and another woman.
This is by patterns in details of her face, vocal inflections, and the tone of her voice repeated by going in and out of the room.
If you presented Aiko with plans of another room and asked her to continue studying the environment without a visual shutdown, kept bringing the plans in and out of view, and asked Aiko to travel that rout over and over and come back to position one, and then asked where the bathroom was, that would be learning, sort of like a cat, dog, or even a blind man.

Motion Trackers

There are motion trackers in her eyes which also twist her neck, but what about when she isn't tracking anything.
Does she actually see at those times?
You know how the old tube TV sets had a reflector in back which can piec together pixles in a negative, and upside down image?
I imagine a human brain is similar before the brain processes it, why we see colored objects when they're really clear, but how does the brain do this?
I hope you can simulate it in the future.

Infection

I've noticed Aiko can't vocally simulate emotions (can be fixed in future models).
Although it sounds trivial, it's important for advertisers and the home consumer to think of her as a real woman.
BRAINS could also be adjusted for spontaneous emotion, so hurting her arm wouldn't just result in "Stop that you are hurting my arm," but an actual sensation in her software (recorded as E.E.G. type signals).

Truth Detectors

Truth detection is now a mind reading technology (memory of facial expressions, sweat censers, etc.).
I believe these bio recordings can be transfered to mechanized software (see Project Epoc), like a video tape transfered to DVD.
This would give Aiko even more of a human quality, and I'm surprised nobody else (to my knowledge) has thought of this.