söndag 14 september 2014

Robotic islands

Retiring at 41 always raises eye brows (alright, very close to 42... and I'm actually still staying on as the CEO and probably won't get off totally and irrevocably until I turn 43). Even people professing to know me have had a hard time wrapping their heads around the idea of not working.

These are people that I am sure have claimed to "hate Mondays" or wanted "summer to never end" or hoped for "just one more week off work". Still, they seem to think that it's more natural to sit at a desk 50 hours a week, shuffling papers, going to meetings and generally "play office", suffering from the sometimes Kafkaesque or Dilbertish dealings that characterizes life in a cubicle maze.

I have so much more I want to do, that work stands in the way of. I want to read, to write, to synthesize information, to teach, to draw, paint, play, compose, research. I want to keep up to date with all kinds of technology; bio, nano, AI, robotics and how they interact and help solve the big issues of water, energy, poverty, famine, longevity.

Take the small topic of robotics for example. I am at the same time amazed by the progress made in specific robotics-related areas, and perplexed by the lack of synergistic collaboration and combination of various solutions to robotic challenges within balance, vision, touch, grip, movement, evolutionary organical algorithmical learning of complex behaviours etc.

  • There are robots with impressive balance, with just one large foot consisting of a big ball
  • There are robots with vison based on ultra sound, others with laser, or stereoscopic vison, or infrared, or magnetic vision
  • There are robots with fingers with pressure sensors
  • There are robots with evolutionary algorithms that enable them to learn to flip pancakes by trial and error
  • There are virtual robots that evolve how to stand, walk, etc from random movements and digital "rewards" for what is considered good behavior
  • There are robots evolving their own language by simply standing in the same room talking to each other
  • There are robots that guess your mood from your pulse, body heat and facial expression
  • There are robots with four legs that can run faster than Usain Bolt
  • There are robots on two feet that can play football even if they can hardly stay upright
  • There are flying robots with several rotor blades that can collaborate to play instruments, to catch balls in a net, to balance vertical rods or just operate a camera following a wind surfer

On top of all these rudimentary and practical solutions there are avenues for constructing artificial muscle tissue, artificial brains mimicing our actual brain architecture, organical evolutionary designs for light and strong materials for new types of joints, limbs and muscles...

I want to see it all come together.

I want a robot with small pressure-sensitive balls for fingers (the same kind of balls the balance bots above use as feet, but much smaller, to enable feeling slippage and movement), at least two "thumbs" per hand and finger joints going both ways, as well as rotating around the axis - the wrist too of course. This should solve the problem of breaking cups or letting them slide out of the robot's grip. Carpenters, car mechanics etc should be consulted regarding desired dexterity. Organic algorithms and virtual testing environments could come up with fantastical new body designs that are more optimal than ours, better then animals, better than insects and spiders, regarding the number of fingers, thumbs, limbs, degrees of freedom etc.

I want the robot to have normal legs that can walk, at least two of them, possibly three or four, as well as telescopic limbs protracting from the back and torso (to stop it from falling and to regain an upright position faster after a push or a tumble). I want the "legs" to have feet consisting of the same type of balance balls, and the toes and heels of smaller balance balls, all pressure sensitive.

I have seen physical bipedal robots that when pushed hard, "try" a hundred possible reactions, simulated virtually, in a fraction of a second, choose the optimal one and place the foot accordingly, thus regaining its balance immediately. Of course, it isn't difficult to program even if it might be difficult to make the motors respond fast enough. The next step is to make the right actuators and motors prepared before the robot is hit, to make forecast what is going to happen, given what the physical laws have to say about the object coming its way.

I want to understand why the open source project of robot platforms isn't catching on. I want to know why, since a lot of the robot research is done with public money, they don't collaborate more on combining various solutions to balance, touch, vision, dexterity and intelligence. I know the robot football teams have to share there inventions with each other but they still don't really seem to get anywhere. OK, that's unfair, of course they do, but it's going sooo slowly.

Here are some examples simply from the top (bottom?) of my head right here and now of things that could be improved dramatically over night:

A lot of current balance issues could be solved by adding balls to the limbs (in particular the feet), adding a virtual trial-and-error algorithm for foot placement, making sure the environment is well-mapped in 3D, when idle, exploring the virtual environment for possible future events like falling or bumping into stuff and making a repository of possible reactions or plans for avoiding such events.

Some vision challenges could be solved by combining ultrasound, laser, infrared, stereo- or quadroscopic light vision. Start with the X-box control system, add sound to map hidden parts of the room in 3D, memorize the room in 3D and the robot's position in the space, update often.

Regarding touch, the robot could use a combination of ball-fingers, pressure sensors, ultrasound, magnets, electricity and laser to determine surface features, friction and density. This shouldn't have to be explored in real time. Most common surfaces should already be known or downloadable. Most objects should already be known and recognizable from a distance and their likely surfaces appriximated beforehand and just refined by the actual "touch" (direct contact or by bouncing light, sound or laser off the surface)

There should be a growing open source cloud repository of solutions, hacks and finished algorithms for identifying and using all ordinary objects, ranging from various fruits and household objects to people, animals, other robots, cars, houses etc.

Nothing above requires new technology, only putting parts together, like when Apple made the iPhone.

And if it doesn't happen soon, maybe I'll have to start pulling the strings myself. Hello Peter, Elon, Ray, Sergey etc.! Is anybody interested in dominating the robotic future together with me (or are you already on the way, in stealth-mode)?

The robots could be used for fun, for status (the new "car"), for work, for companionship, for household duty, for dangerous situations, for entertainment... I just can't believe there would be a lack of commercial applications, if robots got just a little more robust than today.

So, that's just one of my hobby projects going forward, thinking about robotics and how to push the robot evolution. It's probably enough to last a lifetime... but then there is AI, nanotech, biotech, water, energy, food, music, art, entertainment etc., etc. If I am to explore my creativity in music, painting, practical inventions as well as various aspects of the technological evolution from a distance, how could I ever have time over for work?!

I mean, who actually wants to work to make money to be able to pay for things you want to do with the little time and energy you have left? I want to do what I want to do right from the start.

who actually wants to work to make money, to be able to pay for things you want to do with the little time and energy you have left after leaving the office?

Inga kommentarer:

Skicka en kommentar