Well, robbery would be a bit tougher than general mayhem. In the foreseeable future you'd probably need a human in the loop, for example to confirm that the victim actually complied with the order to "put ALL the money in the bag." Still that would remove the perpetrator from the scene of the crime. If there were an open or hackable wi-fi access point nearby it'd be tricky to hunt him down.
This kind of remote controlled drone mediated crime is very feasible now. It wouldn't take much technical savvy to figure out how to mount a shotgun shell on a quadcopter and fly it to a particular victim (if you have one). That's a lot less sophisticated than stuff terrorists do already; anyone with moderate technical aptitude could do it with off-the-shelf components. I'm sure we'll see our first non-state-actor controlled drone assassination in the next couple of years. Or maybe a hacktivist will detonate a party popper on the President or something like that.
Within our lifetime it'll surely be feasible for ordinary hackers to build autonomous systems that could fly into a general area and hunt down a particular victim using facial recognition. People have experimented with facial recognition with SBCs like the Raspberry Pi already.
You can forbid states from doing this all you want, but as technology advances the technology to do this won't be exotic. It'll be commonplace stuff used for work and even recreation.