View Full Version : Ambitious AI / Autonomouns project

12-19-2007, 05:14 PM
printf ("Hello World") ;

My Name is Marcus and I'm living in Northern Norway.
I have for some 20 years been thinking of an "AI" autonomous robotic project, but because of limited avaliability of components that's small and cheap enough for a "hobby-bot", there has been no realization of this project.

A few days ago I browsed arround the net searching for a Linuxbased small controller for another project of mine (neurosurgical technology). Suddenly I realized that these controllers are so powerful that they might be the ticket to fulfill my ancient "hobby-bot" project!

Now I would be very glad to hear some input, so I don't try to invent the wheel again. I'm planning to build a robot and see if I can implement the functions I once thought of.

This is my project outline:


A bi-pedal, humanlike, walking robot with the following features:
* At least 22 DOF
* 3-axial gyro for stability
* Accelerometers in body, feet and hands for balancing and smooth walking.
* 4 pressure sensors in each foot for force distribution and balance.
* Environmental sensors such as temp, light, humidity, pressure etc for general "environmental awareness".


* Artificial vision using colour cameras for face and object recognition / location.
* Artificial hearing / voice recognition and sound locating using stereo microphones.
* Artificial speech using a text-speech engine.
* IR lasers for precision ranging in the sub-mm range located in swivel / tilt -head
* IR lasers in hands and feet for collision avoidance and precision location (pick-up of delicate objects)
* Compass / magnetic field sensor for orienting.

In my original spec the robot has a robotic "twin" for "swarm" / distributed tasksolving. This for experiments like "room-mapping" / orienting / object avoidance / object search (&destroy).:wink:
Now I'm not sure I will build the twin due to overpowering costs, but it would be fantastic to do a thing like this. In case I can build two of thesel they will be put online for user interaction (online robofighting!

The electronics will be built using a "disributed computing" approach. Several on-board, task specific CPU's will be networked for maximum computing power. Eventually a host "mainframe" will be part of the network.

DSP's will be used for video / audio processing and a uCLinux controller will be running the main control logic.
WiFi and BlueTooth will be used for inter-robot networking and intenet / host access.

First of all; Norway isn't a "Meccha" for hobby robots and because of this there is virtually NO local expertise in the field. I have been doing research in the robotics / automation field for years, but only a few days in the "hobby-bot" area.

Now I need some feedback from the hobby community. The feedback I want is foremost suggestions for hardware / software and comments about the project in general.
I have zoomed in on a few products that I'm planning to use in this project:

Chassi: KONDO KHR-1HV or possibly the "BIOLOID" system.
CPU's: fast controllers / DSP's capable of running embedded Linux

The Linux motion controller will probably be "piggy-backed" to the original controller via hi-speed serial line.
One thing I will try to implement is "Autotune" of the walking algorithms. The robot should learn to walk with certain goals like smoooth / non-jerk motion with high progress rate. (therefore need for a lot of sensors, and a fast CPU).


Does this sound like a "healthy" project, or has this been done over and over (in the hobby arena)?

Does the KONDO RCB3 allow for servo manipulation / override through the serial line?
(what I'm thinking of is that the basic motion patterns should be done by the RCB3, but the Linux controller will "orchestrate" the movement pattern. It should also be able to interrupt a motion and take control over the servos (for fine motor adustments during pickup and handling of objects)


Because I can only work part-time due to illness and since all my R&D is privately funded by me and currently goes 100% into my other research, ALL DONATIONS OF EQUIPMENT IS VERY WELCOME!

If you have something lying arround that you don't use anymore, which could be useful in this project; Please feel free to drop me a message! All help will be MUCH appreciated!!!
This could be old servos, actuators, sensors etc, or almost anything useful.

ALL donations will be mentioned on a website dedicated to this project when it's being a reality!

If you represent an importer, distributor or dealer for the KONDO or BIOLOID robots, and donate one (or two) of them, I will personally make you a very special bodymod and "suit" for a this robot. This will be a copy of the suit and bodymod I will make for my own robot, but in another colour. The suit will be very "familiar" to people in the "small-bot" community and the modified robot will be approx 580mm high. :wink: If the donation is substantial, I will give you a prominent, permanent ad at the top of the projects front webpage, (perhaps a complete bot-kit and some small extras), and you will be mentioned at all presentations of the project. You will also be offered to have a "robo-twin" built. (a hardware / software clone).

The robot will be used for demonstrations at the university and promotional items for kickstarting the hobby robotics in Scandinavia, and as such be highly profiled!

Kind regards

Marcus, 70 degrees North

12-19-2007, 10:51 PM

I would start off with a robonova. I think that it is pretty easy to get and easy to modify.

You really only need two gyros and a tilt sensor to make your robot get around. Making the walking smooth I think is done through programming good movements, which can be as smooth or as choppy as you have the ability to program it.

Best of luck getting going.


12-20-2007, 12:35 AM

Thanx for you reply.

Why do you prefer the "Robonova" to the KONDO?

Part of my goal (which is not clearly stated in the brief description above) is to have the robot "learn" to walk by itself. I want to write routines that finetune themselves to walk smooth. Therefore the extra sensors and gyros. My wish for a fast, Linux based controller piggy-backed to the original one is just to be able to "learn" walking by analysing the original controllers performance compared to the readings from the sensors.

Sounds weird to you? Well, I think this is a new appproach to doing things in hobbybots?

Mabye I shold update my first post to include ALL my ideas for the bot? No, think not... I would be locked up in an assylum.. :veryhappy:


This post is printed on 100% recirculated electrons.

12-20-2007, 04:23 PM

I like robonova because there are commercially available double knees and grippers. I am not that handy with tin bashing to make them myself.

I think that you want the robot to learn to walk, basically. But with the use of gyros and a tilt sensor the robot will walk with good code in a smooth fashion. Is your robot going to roam the jungles? No, I do not believe either plateform will enable that kind of robot. I think that a computer that has enough routens available to it, can get by through picking the best option. As well the gyro works independent of a brain so it helps keep the robot stable. There is a good video of two robonovas running on youtube, one with a gyro and one without. The without one falls over and the with one runs great. Mark Bauer got his robonova to jump with the double knees. Not very high, but it did none the less do it. When it lands a gyro goes a 100% of the way in keeping it from falling over.


12-20-2007, 06:24 PM
You wrote
"In my original spec the robot has a robotic "twin" for "swarm" / distributed tasksolving. "
I like your thinking. Why build one when you can have two for twice the price?
You may have a future in Military Procurements.

Until you reach for the stars, you'll never know what it feels like to fall flat on your face.

12-20-2007, 07:27 PM
You wrote
"In my original spec the robot has a robotic "twin" for "swarm" / distributed tasksolving. "
I like your thinking. Why build one when you can have two for twice the price?
You may have a future in Military Procurements.

Until you reach for the stars, you'll never know what it feels like to fall flat on your face.

Hehe.. I know what you mean..

Two at three times the price of one is more likely since the complexity of the project increses dramatically.

My projects usually gets funding or becomes terminated at the early planning state, so I'm not too scared of the costs at the moment. Problem is to get funding NOW when I'm still hyped.

12-20-2007, 08:55 PM
You get ... funding?...
Like, money?

C'mon, if NASA can fund a few moon shots, surely you'd invest with more than 'sweat equity'. Mortgage the house. Sell the car. Stop eating. Unplug all non-essential appliances.

If you build a better robot now, when they finally do take over you'll have friends on the inside.

12-23-2007, 09:24 PM
I've been working on something somewhat similar:


using the Bioloid kit. I put together a few new bus devices - a six axis IMU, and foot pressure sensors.

I'm using a gumstix onboard as a smart wireless bus router, with the brains of the system on my PC.

- Jon