Guy was pushing it against the ground, should have been lifting upwards or pushing on the top thingy. Pick up the robot and throw it in a lake, problem solved. Unless it's a underwater pushing robot. If we keep this up we're screwed when the robot zombies come for us. Anyways, I like it.
Fiiiine. He was pushing it down because it couldn't get enough traction on the gravel path. But with proper traction, the thing is scary powerful. We've accidentally battered open the lab door more than once when it went out of control.
Edit: One more video
A curious little filter I devised right at the end of my college career. It's a combination of orientation-sensitive wavelet filters, which are then recombined for the video. Normally they would be used separately to influence the line prediction across the entire scene, rather than per-pixel as we had done before. The noise level in outdoor environments makes it necessary to pull out the big statistical guns even for simple tasks like identifying painted lines.
The woven meshwork you see in the video is actually individual blades of grass. The size of the filter lets it locate patterns even when there are obstructions, and this is how it interpreted the results. In practice, the response from grass is an order of magnitude lower than painted lines, but it is difficult to represent the gap with an 8-bit spectrum. I probably should have gone with false-color...
Sadly, by the time I got this far, it was too late to implement it. But then, that's always the way it goes on design teams. You lose a lot when the upperclassmen graduate.
Sun Sep 08, 2013 11:49 pm
