Also the extra cost of the accelerometer and gyro is almost negligible these days, thanks to smartphones requiring them to be made in such high volume. For both together, it's maybe $20 or less in volume.
I feel like the biggest issues might be missing ambient spatial awareness for the user - with that non-wideangle lens and mono microphone you can't see the whole environment at once, you have no idea where somebody is talking from, and you cannot turn quickly enough to always track what's happening in a room, if people are leaving/entering etc.. How do you plan to make the user feel more "there"?
Also, micro-managing the movement of the bot would take up a lot of time that I don't have to spend when I'm present locally. E.g. when I'm present myself, I just look somewhere, then think "I want to go there" and it all happens automatically. With the bot, I have to manually navigate around plants and desks. It would be great if we could do the same thing with the bot: just touch on the screen where you want to go, and the bot figures out the rest. Or show a floor plan, touch where I want to go and the bot does it. Do you have plans in this direction?
So... why does it have to be self-balancing? Just add a third wheel and it's always stable - no need for gyros and complicated electronics. And as a by-product, you have immediately lowered the price massively: it's just a few motors then instead of that balancing magic. My price lowering guess would be between 50% and 80% lower. That would be really "the lean startup way" in my book.
(Idea courtesy of http://maddox.xmission.com/c.cgi?u=segway_more_complicated_t...)