Kinect initial test

I've acquired a Kinect sensor, and have done some initial tests on it to see what the depth maps look like.



The quality looks very good, and it appears that there is relatively little noise in the range measurements. As I had expected, this sensor definitely doesn't work outdoors, but under all the indoor situations in which I've tried so far it seems to work well. Just like with stereo vision there is a minimum ranging distance, but this is quite short.

The sensor itself is quite bulky, and this is probably because it contains a motor used to tilt it up and down. If I use this as a replacement for the existing dual stereo camera head on the GROK2 robot then I'll need to design a fairly industrial mounting for it.

With the arrival of sensors like this the quest for good perception on mobile robots operating indoors and which don't require ranging distances more than a few metres is reaching its conclusion, and we are almost in a position where yet another of the necessary criteria required for a new robotics industry to begin can be ticked off of the list. No longer is it the case that to obtain good short distance range data you need to buy laser scanners costing thousands of dollars - like the nodding laser on the PR2 robot. It's also to be expected that other companies will try to produce similar sensors and that their cost will come down still further, as happens with all new consumer gadgets.

The next step is to try using this sensor for SLAM, and I think that Willow Garage are already in the process of doing that. The prospects for being able to build and navigate within 3D models of indoor environments over the next couple of years are better than they ever have been.