AI

[Humanoido]

What does it take to develop some really good AI?

This is a partial list of resources for a Robonova, inclusive of upgrades, that can be useful in generating some AI routines.

Anyone care to add to the list? Did I forget something?
humanoido

-----------------------------------------------------------------------------
Resources for Robonova AI
-----------------------------------------------------------------------------
Motion Control - Servos, Gyros, Accelerometer, Compass
Storage - Memory

Output
Motion, Speech Synthesizer
Sound/Speech/Music Playback
Music, Notes, Tones, Beeps
Peizo Speaker
LED
LCD
Bluetooth, RC
Timer
VR Software
Computer
Battery Sensor
Ports

Input
Vision - Camera, IR Detector, Sonar, Ultrasonic
Recognition - Vision (Face Recognize)
Brightness Level
Sound Level
Sound/Music Recording
IR Receiver
Bluetooth, RC,
Speech Recognition
IR Controller
Timer
Touch Sensor
Microswitch
Hall Effect Sensor
Servo Sensing
A/D
Ports
Computer

Second Controller Board
AI
Recognition, Judgment, Decision, Choice,
Spontenaity, Idea, Reason, Think,
Speculate, Ponder, Daydream,

[Robo1]

IT's a good list.


it just makes you think how big the the task is.

Bren

[DirtyRoboto]

An Amoeba posesses more AI then RN1 and has fewer add ons. Why is this so?
Are we looking at a picture that is too big for us too see or have we missed the point?

An organism needs a drive, what is RN1's drive?
Rn1 suffers due too the fact that humans care for all of its needs. It has no goals in life but to be a toy!

A blind, deaf, mute retard can enjoy life with minimal stimulus, yet we can add many sensors to the RN1 and still never bring it too the level of an blind, deaf, mute retard. Why is this?

[i-Bot]

Marcus, I think you are so so right,

To make the most of limited awareness, and to exploit that is the best.

I don't care if T takes weeks to make an intelligent decision. Timing is easy to fix.

We seem to have lost the plot on the AI topic. The term "self" is ambiguous meaning both a thing, and an ego.

That T is now self aware from a "thing" perspective is true from the force and gyro feedback. Making a big list of AI desirables will not improve this. Totally aware is not possible.

Self (thing) awareness comes first, , environmental awareness next, modelling of scenarios based on these next. Maybe real world success of these scenarios == ego, but probably too early to tell !

If T can climb stairs (especially if out of standard specification!) and get back to main. then he should be the proudest RN on the planet.

RN1's drive is to serve your script, but don't tell him he's a toy ! One day his progeny may be our boss !

[i-Bot]

Marcus, I think you are so so right,

To make the most of limited awareness, and to exploit that is the best.

I don't care if T takes weeks to make an intelligent decision. Timing is easy to fix.

We seem to have lost the plot on the AI topic. The term "self" is ambiguous meaning both a thing, and an ego.

That T is now self aware from a "thing" perspective is true from the force and gyro feedback. Making a big list of AI desirables will not improve this. Totally aware is not possible.

Self (thing) awareness comes first, , environmental awareness next, modelling of scenarios based on these next. Maybe real world success of these scenarios == ego, but probably too early to tell !

If T can climb stairs (especially if out of standard specification!) and get back to main. then he should be the proudest RN on the planet.

RN1's drive is to serve your script, but don't tell him he's a toy ! One day his progeny may be our boss !

[voodoo]

this is one area of robotics that has a big draw on me i agree we need to set rn1 or any other robot a set of goals first then develope ai to achive these goal an example i can think of is some of the biobots you can find on the web they have a simple goal to find light to feed from ( they have solar panals and are hardwired to follow light to constantly recharge there power soruce i have done a lot of reading on this over the years and i am fasinated by the subject

another idea i had for a humaniod robot was in the form of gettin a chatbot program to run on the robot so that you could have a convosation that would pass the TURIN TEST

[Humanoido]

DirtyRoboto: An Amoeba posesses more AI then RN1 and has fewer add ons. Why is this so?

This brings to mind "The Game of Life" by John Conway in 1970.
http://en.wikipedia.org/wiki/Conway's_Game_of_Life
It's a form of artificial intelligence that resides on the computer, defined by rules of a computer program based on some math models.

It is entirely possible that we can develop our own models, and mathematical rules for our Robonovas to designate a form of life.

Another thing that comes to mind is having a self charger. RN has only 1 hour or less of activity and then he dies. The ability to find a feeding station to recharge the batteries should be a priority on the list for autonomy and AI. I don't recall if anyone has threaded a charging station.

"The Turing Test is a proposal for a test of a machine's capability to demonstrate thought. Described by Professor Alan Turing in the 1950 paper "Computing machinery and intelligence," it proceeds as follows: a human judge engages in a natural language conversation with two other parties, one a human and the other a machine; if the judge cannot reliably tell which is which, then the machine is said to pass the test. It is assumed that both the human and the machine try to appear human. In order to keep the test setting simple and universal (to explicitly test the linguistic capability of the machine instead of its ability to render words into audio), the conversation is usually limited to a text-only channel such as a teletype machine as Turing suggested or, more recently, IRC or instant messaging."

In the time of this proposal, numerous claims to this effect have been made. It would be of great benefit to work on such a project for RN.

Indeed, "Swarming" exhibits intelligent behavior to some degree, and it's models are very significant. Since we each have only one RN to work with, perhaps swarming models can apply to internal intellectual thought AI models.

humanoido

[Humanoido]

limor was talking about an AI book shown at the lower right. Then I noticed a section on artificial intelligence there. There's also a book that I posted about which gives sample AI programs written in BASIC that can be easily converted to RoboBASIC. You don't have to add processor boards - just use what RN already has.

humanoido

[Gort]

Videos on how to use Microsoft Robotics Studio
http://msdn2.microsoft.com/en-us/robotics/bb383569.aspx
I think if we could use the Studio for programming the RN, it would make it easier to open source the AI program for everyone to work on.

[Gort]

Looks like you can us microsoft's robot studio to program the KHR-1
http://msdn2.microsoft.com/en-us/library/bb400943.aspx
VB.net is a add on to basic and Robobasic is a add on to basic. Why can we not us VB to program the RN. I think Hitec and Microsoft should get together and make this happen.

[i-Bot]

It is my opinion that Microsoft Robotics Studio is exactly the wrong way to go for AI experiments.

The exciting opportunity is for AI within the RN as an autonomous biped. Most of the examples of using Microsoft Robotics Studio are based on using the robot (Lego, KHR) as set of motors and sensors under remote control of a .Net program. This is great for animatronics and some robotics applications, but not in accord with my ideas of where AI should be going for autonomous robots.

Microsoft have described how the Robotics Studio can be used for autonomous robots, but this requires a processor and operating system capable of executing .NET programs inside the robot.

While there are great opportunities for simulation, gait generation, etc in Robotics Studio, it is very complex, and this is also at odds with the simplicity and relative predicability of the RoboNova. I think as many AI breakthroughs will be made by using limited resources which are well understood by the programmer, as from massive processing power and lots of sensors.

Would I put a gun in the hands of a RoboNova ? probably not. Would you put a gun in the hands of Microsoft Robotics Studio ?

[Gort]

I think Robobasic is the limiting factor. If you don't want a Microsoft product how about Linux and Java. I still think we need some type of exception handling capability within the robot to log information about run time errors. Did a input cause the error or was it a problem inside of the code. Can Robobasic do these? I have not yet seen this in any manual. Can a try catch type statement be written in Robobasic. Can the Rn store the values in a file so the programmer can look at them latter? I think you need this capability and more that I do not think that Robobasic can give you.

[Gort]

There is a Poke command in the robobasic manual. Which is a write to internal memory. The two questions are, can it be used to create a exception file with enough information to be helpful and two is the RN's internal memory large enough for a file? That is where a 2 gig flash memory add on would be nice. There is also a Print command. That outputs characters to a LCD module maybe that can be used to output to a Flash memory module?

[Gort]

I think it would help to add an external read to a SQL table in the AI program. You can store pictures and voice recordings to a SQL field plus other useful information. I can find no command in Robobasic to do this. I think it would take a lot of bandwidth to send this information back an forth between the RN and a AI program on a PC. The Robobasic manual says that the robobasic commands have been added to the general basic programming language. Can we use all the Basic commands not in the Robobasic manual? That would help in programming the AI. The basic language has evolved into Visual Basic which is a OOP language. If the RN could be upgraded to use a Robobasic version of VB then that could go a long way to solving many of the problem that Robobasic has.

[Humanoido]

RoboBASIC has a complete command list in the manual. It's a very good subset of the full Dartmouth Extended BASIC, with many new commands for motion control and port control. If you are looking for another function not in the command list, it may be possible to simulate it with existing RoboBASIC commands and functions. There are numerous web sites devoted to creating these extended functions, particularly mathematical models and functions.

Indeed, AI processing of photos is intensive, and a dedicated processor would be helpful. RoboBASIC could still do the calls for these routines amidst various motion control programs. I'm thinking - wireless raw data stream to a computer for heavy processing, or one add-on controller with Open Source Architecture so everyone can use it and contribute to its development.

From what I'm reading in the forums, no one can agree on which language to use. Its like one language is good for this but not that, and another language is good for that but not this. Plus, one person likes this but not that, and another person likes that but not this. Or one language has a very steep learning curve (i.e. years) and another does not.

One thing that can be developed is MLPs, or multiple language programs. I frequently resort to MLPs in my programming, often running a simple language that is efficient, and calling in other languages for specific functions. For example, I mixed VB, C++, and B for a cool program that has visual processing, KB processing, CAI - centralized artificial intelligence, and encoders/decoders. So I know it can work successfully.

humanoido