Bayesian Classifier Answers the Question: "Is it art?"

The title is a joke, by the way. Although I would be interested to see what the results of such an exercise would be. "Statue of David... *beep*: ART!", "Justin Beiber's Music... *beep* NOT_ART!".

Almost as exciting as my 0-R Spam filter. Catches 100% of spam with a 97% accuracy rate! Oh mercy.

The reason that I mention Bayesian Classifiers is because I wanted to talk about machine learning. This is the other branch of artificial intelligence and what most people think of when the topic of Skynet is brought up.

Fear not though. Bayes rule, decision trees and rule based learning are actually pretty mild. They are simply statistical methods of attempting to classify data by using the results of previous observations. Mostly harmless.

However, today I'm going to talk about genetic algorithms.

A genetic algorithm is an abstract representation of a mathematical function. They can take many forms, such as a string of bits which might indicate the presence/absence of a set of inputs, or a literal mathematical function "y = cos(x) - 2*z". The range of variables which is represented by the function is called a genome.

This can get a bit hard to visualize, so I often just settle for imagining genomes as Taylor polynomials. Therefore, a single genome consists of "x = A*input1^a + B*input2^b ..." where the values of A and a can take any real number. If some of the inputs are simply a higher derivatives of other inputs, then any arbitrary function can be represented in this way. There is also a rather nice representation involving trees.

If we start off with a population of individuals with random valued variables in their genomes, then we can evaluate each function to see how well it 'fits' a set of training data. The individuals which produce the minimum mean squared error for the training data are declared the 'fittest', and are allowed to survive into the next generation.

This is where the 'genetic' part comes in. There are many ways of 'evolving', 'mutating' and 'breeding' individuals, but the easiest to understand is the asexual method. This means that all individuals except the best performer are killed off (ie, deleted) and then their places are taken by the offspring of the remaining individual. However, tiny random 'mutations' are introduced to each of the new individuals variables - such as doubling/halving the values of A or B, or incrementing/decrementing a or b.

Anyway, thats the 30 second version of genetic algorithms. They can be used to find a semi-optimal solution to many problems, provided you can throw enough generations at them. I have been working with a C++ implementation called GAlib. If you are interested, I highly recommend going through the examples.

Now, some of you may be wondering what all this has to do with robots (actually, most of you are probably already filling in the blanks and peeing your pants in terror).

I've spoken on several occasions about using Robobob as a platform to investigate dynamic balance and movement. I plan to represent the control state of the robot as a search tree, with the robot beginning at a starting node/state and attempting to plan a path of control actions to reach a goal node/state. To navigate the tree, I want to implement a greedy search heuristic which will choose which control actions are most likely to lead to the goal state.

Now here's the tricky part - I intend to implement the heuristic as an evolutionary algorithm which can then be rewarded or punished depending on the outcome of executing the control path on the real robot. ie, if the heuristic gets stuck or can't find the goal state, it will be disfavored whereas successfully reaching the goal will be favored. After a series of generations, I will be able to study the path planning method which has evolved from this process.

Cool? I hope so.

Terrifying? Definitely.

Tremble before DISHZOR!

People are often surprised them when I tell them that my favourite robot is the dishwasher. Not Megatron or ASIMO or the roomba - just your ordinary, garden variety dishwasher.

You see, there isn't much of a consensus regarding what defines a robot as a robot. Some people think they need to be humanoid. Others feel that robots should include sensors and actuators. Computers and software are often mentioned.

One of the broader, but generally well received definitions is that a robot is a machine of some sort, capable of performing tasks on it's own. Begrudgingly, most people will admit that this would include dishwashers. And washing machines, printers and even modern cooking appliances.

Thats right - robots have already invaded your home and you didn't even realise! Insidious!

I think one of the reasons that these everyday robots have slipped under our radar is that we don't have such high expectations of them. We assume robots should be smart, since they are capable of performing the same tasks which humans can - but we tend to assume that tasks which are difficult for us are difficult for everyone.

Sci-Fi has led us to bestow a level of expectation on a robot which is based on their appearance. For example, dog shaped robots tend to be about as smart as actual dogs, despite the fact that their positronic brains could make them just as smart as a human. Robotic bugs will behave exactly like real bugs would, even though they have no reason to seek dark places or put on threatening displays.

In the case of dishwashers, we tend to assume that machines shaped like large bricks will have an equivalent IQ. That's probably why I find this so much less impressive than it actually is.



It's no small feat, but it just seems that a multi robot system like that should be capable of so much more.

This phenomenon also means that as soon as you give your robot a humanoid form, people suddenly have much higher expectations. I think this is why modern demonstration robots are given infant-like designs - if the robot reminds people of a five year old, then they will be that much more impressed when it walks without falling over or grasping objects.



So if you are ever trying to demonstrate how smart your robot artificial intelligence is, remember to make it look stupid so that people will be extra impressed. It also helps to avoid the uncanny valley.

Two steps forward...

I did a lot of work today, improving the robobuilder motion system and writing some custom functions to take advantage of the special features that the wCKs have to offer.

Unfortunately, my efforts were met with a minor setback when I managed to strip another gear - this time in the poor robot's knee. Unlike the first time, or the time after that, the gear which needed replacing was the first gear in the set - the one which is driven directly by the motor.

This leads me to believe that it broke simply due to regular wear and tear - not because of any specific negligence on my behalf.

Gears tend to be under more stress the further away from the motor they are, because they rotate slower, but with more torque. This is why (I assume), I didn't receive any replacement #1 gears with my kit.

Therefore my plans have been put on hold until I can get my hands on a replacement. This should serve as another important selection criteria for servos - metal gears are worth the extra money.

So today's post will not, as I had intended, consist of more awesome videos of my robot doing awesome stuff.

Instead, please enjoy a variety of other people's robot's doing other people's stuffs.



Above is an earlier version of HINA, doing demonstrations for the robo-one competition. I'm fairly sure that the bottle of water is full, which means that there is quite a lot of power in that little robot. As usual, mujaki has put a lot of effort into the custom body parts and it looks absolutely stunning.

In the not too distant future, I'd like to have a go at customising my servos - particularly given the problems I'm having with these plastic gears.



From cute to creepy.

This is phoenix, a hexapod with an amazingly lifelike gait and behaviour. It looks like a real insect, especially when it goes into attack mode!

Phoenix is able to move so fluidly due to an inverse kinematics engine (another interest of mine). This enables fine control over the body and individual limbs of the hexapod, without compromising the balance or posture of the rest of the robot.

When I can get this servo fixed, I'll upload some more videos of my own - please bear with me. If anyone knows where I can get replacement wCK gears (preferably metal) in Australia - let me know.