![[personal profile]](https://www.dreamwidth.org/img/silk/identity/user.png){kind=small}
One of the most interesting things I learnt in the first few years of my artificial intelligence degree was that intelligence can be present in a design in many different ways.
The example given by the lecturer ran as such: imagine you were trying to build a robot horse. The ground it has to deal with is pitted with burrow holes — how do you stop the robot tripping? One way would be to develop a sophisticated vision system which can recognise holes from solid ground, and a complex movement system to dodge them. Or you could just give the robot bigger feet! A small amount of intelligence from the designer can save a huge amount of processing power that would otherwise be required by the entity itself.
The lecturer turned up at the beginning of the next lecture with a box-like vehicle and a grin on his face. Setting the vehicle down, he demonstrated how it would proceed forwards until it hit an obstruction, at which point it would reverse and continue on in another direction. "What's under the cover of this vehicle?" he asked the audience. Suggestions involving electronics and switches were posited. The lecturer then lifted the cover to reveal an incredibly simple and purely mechanical system. The vehicle switches from driving the back wheels to steering the front wheels through mere friction. Just one example of how complex behaviours can be the result of surprisingly simple control systems.
A tutorial exercise at around that time involved the design of a (hypothetical) robotic vacuum cleaner for the university building. What challenges would be present in the environment, and how can the robot avoid them? We were allowed to design intelligence into the control system, into the robot's body, or into the environment itself. Thus making the robot perfectly circular helps it not get caught on chair legs and the like. But how does it avoid falling down stairs? My best suggestion was a magnetised strip of tape stuck across the top of the stairs — easy to detect, and relatively unobtrusive to humans. But this wasn't as nice as I wanted it to be... I concluded that there simply was no elegant way to avoid drops. (I notice the Roomba uses infrared... also not hugely elegant.)
You can imagine then that I was quite surprised when I recently walked into a toy shop to see a toy car driving around on a tabletop, turning sharply whenever it reached the edge. There was nothing obviously special about the tabletop, and the toy was an obviously cheap plastic thing. Naturally, I flipped the car over to see how it worked. The solution was laughably simple: a fifth wheel at ninety degrees to the others, slightly recessed. As soon as the front wheels drop off the edge, the fifth wheel hits the ground, and the car makes a swift turn. Now why didn't I think of that? 8^)
The example given by the lecturer ran as such: imagine you were trying to build a robot horse. The ground it has to deal with is pitted with burrow holes — how do you stop the robot tripping? One way would be to develop a sophisticated vision system which can recognise holes from solid ground, and a complex movement system to dodge them. Or you could just give the robot bigger feet! A small amount of intelligence from the designer can save a huge amount of processing power that would otherwise be required by the entity itself.
The lecturer turned up at the beginning of the next lecture with a box-like vehicle and a grin on his face. Setting the vehicle down, he demonstrated how it would proceed forwards until it hit an obstruction, at which point it would reverse and continue on in another direction. "What's under the cover of this vehicle?" he asked the audience. Suggestions involving electronics and switches were posited. The lecturer then lifted the cover to reveal an incredibly simple and purely mechanical system. The vehicle switches from driving the back wheels to steering the front wheels through mere friction. Just one example of how complex behaviours can be the result of surprisingly simple control systems.
A tutorial exercise at around that time involved the design of a (hypothetical) robotic vacuum cleaner for the university building. What challenges would be present in the environment, and how can the robot avoid them? We were allowed to design intelligence into the control system, into the robot's body, or into the environment itself. Thus making the robot perfectly circular helps it not get caught on chair legs and the like. But how does it avoid falling down stairs? My best suggestion was a magnetised strip of tape stuck across the top of the stairs — easy to detect, and relatively unobtrusive to humans. But this wasn't as nice as I wanted it to be... I concluded that there simply was no elegant way to avoid drops. (I notice the Roomba uses infrared... also not hugely elegant.)
You can imagine then that I was quite surprised when I recently walked into a toy shop to see a toy car driving around on a tabletop, turning sharply whenever it reached the edge. There was nothing obviously special about the tabletop, and the toy was an obviously cheap plastic thing. Naturally, I flipped the car over to see how it worked. The solution was laughably simple: a fifth wheel at ninety degrees to the others, slightly recessed. As soon as the front wheels drop off the edge, the fifth wheel hits the ground, and the car makes a swift turn. Now why didn't I think of that? 8^)
Tags:
![[identity profile]](https://www.dreamwidth.org/img/silk/identity/openid.png){kind=small}
From:
no subject