Slashdot Mirror

It's a hard problem, I've also worked on for the last several years. You're combining research problems in AI, computer vision, localization/mapping, motion planning, human interaction, etc.; each of which demands high end hardware to run its computations, but then you want to do it on mobile platforms with tight constraints on power and sensors.

Then in order to modularize things you have to come up with a generic interface for each piece in order to abstract it. I think this aspect in particular kills reusability, because these pieces are all so interdependent. Each module needs internal state from the other modules to interpret its own data, and depending on the implementation used for each module and the actual robot hardware it's running on, some types of data may or may not be available, and some outputs may or may not be possible. It's a combinatorial explosion of different capabilities, which leads many people to write to their current hardware and their own specific implementations.

I entirely agree to make progress we need to address this issue. Asking every researcher to reinvent the wheel in all of the related fields before they can work on their own piece is ludicrous. And it doesn't help that many implementations are very sensitive to robot specific parameters, so even if a research publishes his code for a problem (which IMHO should be part-and-parcel of publishing results), you might still have a hard time running it on different hardware where sensor or motor models differ or may not even apply.

You know, Sony is supporting development of new software for the AIBO (*) You can create new personalities using high level tools like R-Code, or download their software development kit (for free!) and code in C++. It's an impressive piece of hardware.

If you're interested in the low level processing, which allows direct processing of the camera images, networking support, real-time control of joints, etc., then of course I'm going to recommend the software framework I'm currently working on: Tekkotsu. However, there are number of other options available as well. (see my prior post)

Anyway, we'd always like to have more developers -- help show Sony there are advantages to opening their source code!

(* or at least now they are, originally they didn't like the idea so much, but they seem to have gotten over it somewhat -- hardware is still tightly locked down, but the software interface is pretty available)

It's easy to write code for an AIBO... there's a number of open source software frameworks for this great hardware platform -- although of course my favorite would be my own: Tekkotsu

And it's all supported by Sony -- no hacking required!

There's a variety of levels you can code at as well -- there's several high-level scripting languages like URBI, R-Code, and even a couple upcoming Python interfaces, as well as a number of low-level C/C++ interfaces (e.g. Tekkotsu) which can run onboard and directly process every bit and byte, or remote control from your PC for maximum horsepower.

It should be noted that our web page is being updated, and some of the links seem to be going to place holders. Sorry about that.

And if you really really want a joystick, our robotics team hacked joystick support into tekkotsu mon in about 30 min. It turns out that joystick gets annoying when you need the dog to clear a long hallway, and the mouse really is a better input device, but its still an easy modification.

System SDK: OPEN-R

or a framework which builds on top of the SDK: Tekkotsu

The goal of the Tekkotsu project is to eventually be able to train a robot like a dog.
It's open-source, so you can try your hand at it if you think it should be easy ;)

--Tom

For those of you who have a 210 or 220, some guys at CMU have made a framework for programming the aibo that is much easier to use than OPEN-R. The website is at www.tekkotsu.org
We are using it to program the dogs to perform urban search and rescue, and we have been very happy with Tekkotsu in general. Its worth checking out if you want to start programming the aibos quickly and still have power to do what you want.

As previously reported on Slashdot, Sony has opened up the API for the AIBO, and it can be programmed in C++. My lab has created an application development framework for the AIBO called Tekkotsu that we think people will find helpful. The code is open source and GPLed. Visit Tekkotsu.org for an overview, downloads, demos, and documentation.

For those of you who would like to do research with real world data, but don't want to spend time actually building the robot before you can use it - the Aibo is a great platform.

There's a free software development kit from Sony as well as an application framework to get you jump-started.

It once cost a substantial amount of money to buy an AIBO devkit from Sony; I now think they're wising up by opening up the SDK and allowing GPLd libraries like Tekkotsu to exist.

I program AIBOs at school now. AIBO already walks, of course, and it has an accelerometer that may be able to deterimine when it's about to slip off your back. Combining them both is a simple matter of programming. For all the trouble it gives me, it'd be nice for AIBO to give a little back in return.

Once things get less busy (i.e. perhaps in the summer), I may try coding this up. Shouldn't be too difficult.

BTW: if anyone has an AIBO and is frustrated with trying to program it, check out our new development framework. If I ever do get that massaging AIBO to work, maybe mention of it will turn up on that website.

--Tom