I'm skeptical that a stationary ball lying on the floor (or rolling into a hole) would provide better information. Of course, it's better than not having that.
How do you make the helicopter rotors so that they don't break the first time they hit something and also don't cut people's faces up. In many situations, have a flying helicopter in a small space can be much more dangerous than a ball on the floor.
The cost of RC helicopters is quite staggeringly low, and they are usually possible to repair. The blades can be painful if you fly into people, but I doubt people appreciate having balls land on them either. At least with an RC you could potentially see where you were heading and stop. If the use scenario is to send these in to places too dangerous to send people, I don't see much issue.
I don't think autonomous flying is necessary, but there has been research done to show it's possible (but as you say, not trivial).
I'm still unsure if a ball is going to work well, or whether it's just going to roll under a desk.
It's also not advisable to assume that the users would be highly trained beforehand. A major benefit of a device like this is the fact that it is a very simple unit that is easy to deploy to get a quick and easy view of the environment around a corner or at the end of a tunnel.
Additionally in fires you have a lot of air drafts from the heat, making controlling an RC device even more challenging.
Surely even at $500/ball, it'd still be cheaper and safer to use a system that requires close to no training. From the demo video, it seems that the ball also detects carbon monoxide - I imagine detecting other chemicals would come in extremely useful as well.
At a glance, it seems like there's a lot less that could go wrong, given the number of moving parts involved with the helicopter.
Throwing a ball is simple. Flying an RC helicopter is difficult.