In December, Orbism will be moving office, and one of the changes to go with the office change is that Orbism will have it's own little RobotWars groups. In the beginning, the robots will likely be limited in size and weight, but eventually, we plan to have a working monster of a thing which will bash nine shades of hell out of all contenders.

My own plans for my own robot include a certain amount of artificial intelligence, so the robot can guess what is the best action to take, and start on it before I tell it to.

Neural Networks will play a large part in the artificial intelligence, as I am currently trying to understand them.


Navigation. In order to navigate properly, the robot should know where it is. This will be accomplished through a small series of programs.

First, an image of the robot's surroundings is taken from the two (at least) cameras on the robot. This will result in two slightly different images, as they are taken from two different points.

Then, the images are processed by a neural net to find common points, and build up a 3D image of what the cameras see.

Then an algorithm reduces the data by finding lines or planes that can replace groups of dots.

The shape data can be refined further by grouping each plane and line with others to form 3D objects.

The robot's relative position can be obtained by doing some triangulation based on the larger 3D objects, which will include floor, walls, and ceiling.

Additional data can be gained from this process as well - any 3D objects which have changed position in the last few iterations of the position calculation can be assumed to belong to something to avoid, or at least be wary of.

Please send any comments about these pages to kverens@contactjuggling.org
Valid XHTML 1.0! Valid CSS