Another test of the visual memory.

After Truckbot wandered around the uni it had a much larger map to display; and the s/w used to display it results in even more mess than the prev test.

Again the clip shows the original raw video followed by a "scan" of the visual memory.

The final display is a montage of all the images it's seen overlayed on a large display. The "map" is so big (20Kb x 20Kb) it can't been visualised in one svga image, so I've made a small program do a simple "maze run" from the robot's start position to its final position, displaying the visual memory as it goes.

The result shows e.g. all the joints in the concrete path where the robot has been.

The "memory" at locations corresponding to severe jolts (e.g. Truckbot ran off a concrete kerb at one point and it required some quick radio control to navigate one doorway at another point) shows a "confused" image. But the memory contains internally complete images that can later be examined -- it's the scheme I'm using for visualisation that's causing the mess. :) (less info)