The DNA of robots

Let's try a simple exercise of fancy graphical representation of a large number of multivariable vectors, in our case, robots in history, if only just because we love Visual Complexity. We can assume that fictional robots can be included in our timeline. After all, what would robotics be without people like Asimov?

Naturally, a timeline needs to be represented in function of time. However, rather than using a lineal time scale, given the much larger concentration of events in the last 50 years, we can go for a log one, where older events are squeezed in a time slot that may go for centuries and still we may have the same space for a single, very busy year. Basically, the idea is go logarithmic, and compress sparse data into short lengths while we keep dense areas well spaced.

Now, we need to represent robots, but quantify somehow its nature. We decided to do so via a MySQL database that we could modify via web. How do we define a robot, though?. Smartness is not the universal answer: if you ask someone at random who is smarter, Spirit or stair-climbing P3, you might be surprised at how many people think that any 3 years old kid can do the last, yet she most likely will never travel to Mars. Furthermore, if someone asked me if I'd rather have Kismet or Mazinger Z, hey, count me in for the big, dumb, fighting thing!

A major discussion at our labs -that was settled by sheer force of stubborness and endless changes just after everything was already programmed- led us to assume that robots could be characterized via four features: i) smartness, where 0 corresponds to teleoperated robots and 100 to human thinking1; ii) adaptability, where 0 corresponds to non-moving robots in predefined environments and 100 would be ... no, not an inhabited planet, but a metro station at rush hour!; iii) biomimicry, where 0 corresponds to robots that have no similarity whatsoever with a biological being and 100 corresponds to robot that look exactly like an animal -humans, of course, included-; iv) manufacturing, where 0 corresponds to robots whose creator had to model every little piece and 100 to commercial models. To avoid further confussion, we also added a "reality" value that most of us thought should be a Boolean (0=fictional robot/1=real one), but finished as an integer ranging from 0 to 100 because someone thought -very loud- that it could be controversial not to have reality-challenged things.

Fixing a parameter value was also pretty tricky: if you ask 10 different people how smart a robot is, you get 10 different answers. This was solved quite democratically: we enabled a vote system where people opinions on the same robot would be averaged2.

Now, any robot could be represented by a 5 feature vector in time, but, unfortunately, we humans are not very good at visually appreciating 6 dimensional spaces as such. Hence, we did a bit of a trick: we equaled the first four vector elements to C,M,Y and K components in the CMYK color space. The fiction component was assigned to the Alpha channel, e.g. transparency, so that real robots seem more solid than fictional ones. Then, we drew a sphere for each robot, whose color and transparency was related to its nature.

For example, we could state that a hobby tracer mesobot built by a student team is (15,25,0,25), e.g. solid pale violet, if we assume that it is not biologically inspired, its environment will not change much, kids designed it using commercial motors and a microcontroller and it does not have a killing strategy capable of evolving beyond human line-tracking imagination. An ancient automata would be instead completely yellow (0,0,100,0), because it would be done from scratch, biologically inspired like your average kouros and mechanically preprogrammed to repeat the same set of actions over and over. If we do not want spheres to overlap, though, we need to distribute them in a cloud around the year they were created in. And, hey, here comes a new variable in play: size. If just for the word play, we decided that size would be importance: we Googled ("name of the robot" \& "robot") and scaled items according to the number of results. Thus, Soujourner would be larger than, say, RUR, but significantly smaller than Wall-e.

If we check colors through the database, distribution is strongly discontinuous, and there is no noticeable trend, except at the beginning, where robots were all different shades of yellow: bright and solid for real devices and darker and more transparent, for imaginary ones. Indeed, a line is not the best dimension to show interdependencies, so we discussed once more and someone suggested a circle. In a circle, similar robots could arrange themselves into groups and see if there are interdependencies among them. This arrangement was performed via a electrostatic field set of equations, where balls attract or repel each other depending on their distance and color likeness. Some heurisrical parameter fixing work led to field convergency after some iterations and, indeed, groups become now more clear. However, circles do not give sensation of time, so they do not represent trends. Instead, they rather represent types of robots that have consistently appeared again and again in human history, i.e. fictional thinking humanoids.

To get a sense of time evolution, we decided to just up the circle in 3D and not only it did the trick but now the whole thing looked like a DNA helix. Plus everything could be represented in OpenGL and visualized in the web via a 3DXML plugin. This representation presents several interesting features. First, similar robots that are close in time conform groups of balls on a side of the helix: the larger the cluster, the more important this robot tendency has been.

Homogeneous groups at time origin extend to whole rings of the helix, when robots were all automata and ranged only from yellow to green. However, as soon as robots diversified, different trends occupied different sides of the helix that could last for long periods of time -like the aforementioned fiction humanoids or robotic arms- or just be locally punctual, like flying robots before the XXth century. In any case, close clusters would correspond to robots similar in color and, hence, in nature, so transitions between them should be smooth unless something made evolution jump abruptly. Such changes might be correlated -even with some delay- with important events in history, like marxism at the beginning of XXth century or massive commercialization of cheap computers in the 80s. Trends are very clear at the bottom of the helix: yellow clusters of automata and green clusters of fictional androids all the way up to the XXth century. After 2000, though, the helix is so packed with robots that it is not easy to detect trends anymore, yet, there are noticeable streaks of yellow, green, cyann, pink, red and blue, corresponding roughly to humanoids, both real and fictional, rovers, medical robots, toy robots and assistive devices.

1This could be argued, sure, but let us assume that at the moment ANY human is smarter than your average bot
2And, in case of disagreement, us, all-mighty system administrators, could reset the robot to our favorite value in a really democratical way.


Post a Comment

Newer Post Older Post Home
There was an error in this gadget

Recent News

-Biometrically adapted wheelchair control paper accepted in IEEE Trans. on NSRE :) -New paper on collaborative navigation in hospitals accepted in Autonomous Robots
There was an error in this gadget


Recent Comments