the story of robocology

robotics is an emerging discipline. or not. man has strived to create autonomous machines and we've moved from mechanical automatons to electrically controlled machines running pon computers with fantastically complex and long algorithms. we thought that with our newfound discipline of electronics and its rapid advances, we could soon create thinking, autonomous machines with ease. And we quickly ran head-on into a brickwall. moving around a real-world environment filled with obstacles in the way is something we do in our daily lives without even realising what it takes our brains to accomplish this task...until we try to make computers armed with video cameras and sitting on wheeled platforms do this. I believe we're still having some difficulty.

the academic fraternity that deals with AI is coming up with many new computer programs and algorithms for visual processing, moving around obstacles, having robots get up on their feet and walk properly. Many ideas and schools of thought. Then there are those who prefer to start on a lower level: try achieving the intelligence of insects and other 'simpler' organisms before we try to create artificial humans. Years ago i came across this idea as i read one of my first encounters with literature on robotics research (a book approximately titled "Behaviour-based robotics") Then, i felt this was quite lame and unnecessary, but i now think this is the way we should proceed. I'll first have to set the background and explain the rationale for a methodology of artificial intelligence research that i will also outline here.

The various physical and mental abilities of living creatures today have evolved over time, to meet the demand for survival. And survival is for reproduction, which is in turn for the continued existence of the genes. Creatures own the capabilities they possess to allow them to survive in the ecosystem they call home. It was natural circumstances which forced creatures to become smart enough to find their way around based on the visual information from their eyes, to execute the hunting of their prey...Perhaps by analogy we can try to reverse-engineer this process of evolution by which nature gave living creatures their intelligence. Perhaps we could try to create robotic creatures that similarly have to fight for their survival in a robotic ecosystem. Then we'd put ourselves in the shoes of mother nature...how could we engineer these robots to give them abilities which they need to survive, just like their biological counterparts? Using the 'simpler' insects as inspiration, thus removing other distractions like visual processing by advanced eyes that mammals possess and social skills like language etc. we concentrate instead on creating solid, robust control mechanisms for 'simpler' tasks insects are capable of and then building upwards to higher level skills which are more difficult to design.

First and foremost, this approach serves to guide us and give us focus in our quest to develop artificial, intelligent systems.
The second reason for this bottom-up approach is related my belief in how to 'create' intelligence. (i don't think "create" is a very appropriate in this case, but i couldn't find a better word)

I disagree with the approach of trying to programme our way to intelligence. that somehow we will one day be able to develop an algorithm that would make a robot intelligent. besides, many of the efforts in software are directed towards specific "intelligence goals", such as visual obstacle avoidance, recognition of facial social ques, locomotion, moving a hand to manipulate something, working as a group...I think we need a common underlying methodology to achieve all these. I would refer to the brain as an example of this. From insect brains to human brains, organic brains (yikes that sounds awful for some reason) are composed of complex and vast networks of neurones. Neurones and the sound interconnection of neurones is the basic methodology by which nature builds its brains and creates intelligent creatures. Using these basics, nature has shown us what complex and interesting and intelligent behaviours it can create. And for that reason i think we should give nature's method a try too.

This approach of creating a robotic ecosystem and starting out by creating the simplest and most primitive robotic creatures which possess the most basic skills that nature's most primitive creatures have facilitates our playtime with creating networks composed of basic functional units and once we can get our own networks and robots to achieve the basics as exemplified by nature's primitive creatures, we can build on our success and use what we have learnt to get there, to create more complex creatures.

Mark Tilden's BEAM robotics is a well known effort that uses this approach. But the robots so far, while exhibiting interesting behaviours, are still purely reactive devices. No memory, no learning, and the networks cannot modify themselves, their connections. The last feature is important, and in neuroscience, the ability of the brain to do this is called neuroplasticity. Plasticity. Important quality. That'll be a gap we can attempt to fill. What matters in a cognitive system that's made up of networks of functional units is its architecture. Methods and ways of combining the units to create behaviours is the goal for this project. And we hope to progress from the simplest behaviours and skills in the most primitive organisms to the most complex ones.


More details of the robotic ecosystem in the next post ;P