Enabling the Kinect V2 for PR2 has been a little tricky. The very high bandwidth of the sensor (almost fully saturating USB 3.0) combined with the high frame-rate of the sensor means that a dedicated computer is usually needed for data processing. Luckily the PR2 team at Clearpath is here to help!
Comparing the first generation Kinect to the V2:
|0.8 – 4
|0.8 – ~10
|57 deg (h) 43 deg (v)
|70 deg (h) 60 deg (v)
|320 x 240
|512 x 424
|640 x 480
|1920 x 1080
Another interesting quality perspective is that the Kinect V2 takes 300 depth samples per second, and then sub-samples the data to 30 Hz. This results in much more reliable data. The Kinect V2 also is much less susceptible to interference from lighting, as it uses a different approach to depth detection. Thus changes in light and even some incidental sunlight are ignored by the system. Lastly, we have found that the actual range of the Kinect V2 to be much greater than the Kinect.
For our team here, working with the new Kinect was fairly tricky. The large amount of data, combined with the requirement of having a graphics card and USB 3.0 support made integrating the PR2 Kinect 2 backpack quite a challenge. First the system consumes a lot of power, so a 18V – 72V regulator had to be installed inside the backpack to provide power to the systems. Also, fitting in a full powered graphics card and high end computer into a reasonable size involved some careful design. Our super-powered PR2 hardware engineer Bryce designed some great looking mounting options for the backpack.
Image and depth information generated by the Kinect V2 is processed by the backpack to create point clouds. These pointclouds and all other data is then published to the ROS master at the C1 computer on the PR2.
A huge thank you to Thiemo Wiedemeyer for developing the adapted Kinect V2 driver, which this add-on depends on! We will hopefully be putting in some time to make the package even better by transferring to nodelets sometime soon.