I think the donut shape is actually pretty clever for this. As stated by the article it has camera(s), and my bet is that it has enough of them that no matter how you throw that thing around, it always has 360 visibility around it. If the cameras were pointed partially towards the inside of the donut you'd also get some valuable depth information derived from having overlap with adjacent cameras...

Same goes for the microphones and speakers, place at least 3 of them evenly around the donut for some nice beamforming.