If you were hoping to enter the 2013 Micromouse competition then the sad news is that it's all over, but there are some interesting videos of what happened.
The 2013 USA Micromouse competition was held in Long Beach (CA) on March 18th. The task was simple enough - build a small autonomous robot that can run a maze as fast as possible. The prizes were small - $500 for first place - but the honor of this long running AI competition is great. It all started back in the 1970 and today there are a number of similar competitions held all over the world.
When you look at the contest you might think that all that you are seeing is some remote controlled toy being guided though a maze or a "wall banger" algorithm that gets the mouse through by brute force alone. However, if you watch for a few moments you will notice that the mice seem to know where they want to go - this is more subtle than you might think.
The mouse first has to map the maze using a suitable search and optimization algorithm. As well as simply finding a route to the center, the mouse usually explores the maze further to see if the route can be improved. After working out a route the mouse is set to re-run the maze as fast as possible.
The search algorithms used vary in sophistication from simple A* to dynamic programming but over the years the sad fact is that it is the fastest mouse that generally wins rather than the smartest. To make the task more computationally interesting a new version of the competition has recently been introduced with a 32x32 maze rather than the classic 16x16.
Notice also that the maze places a physical size constraint on the mouse so the winner has to be small, fast and intelligent.
Hulk 2 - Picture MicromouseUSA
First place this year went to Hulk 2 from the Institute of Technical Education (Singapore), second place went to WaTiu from Lunghwa University of Science and Technology (Taiwan) and the third place was Excel-8b from the Institute of Technical Education (Singapore).
We have recently seen how much neural networks can learn, but the latest example is very strange. Google has succeeded in teaching a neural network to be a computer simply by showing it programs and w [ ... ]