For me I would _want_ the data to be sent in parallel to some base station for logging, debugging, etc..
And I remember that the display of the map in some video seemed real-time.

There are developement environments that faciliate this type of partly parallel, partly distributed programming.

You can build a map inside the flying controller and duplicate the calculus in a stationary one, to have “broadband access” for display and analysis, what the current state is in the mobile device. You just have to define some interface that gets Y-splitted, one copy locally processed, one remotely transmitted and processed on a PC using the same algorithms.
It depends on what you want to transmit, raw sensor data or to some degree processed data, how much bandwidth you have to use.

For learning purposes sufficien I’ve seen such a setup even for Lego NXT using Lejos.

My drive is towards the ASUS Atom with Nvidia ION platform, where one could use ~ 50 shaders for calculus. But this will still need to much power for a quad this size. (So I will begin with somethng ground-based, perhaps later a blimp

Have a look at readily-built quad or octocopters, but insist on a remote control unit (in your hands _and_ the counterpart in the copter) that has already a built-in serial interface. This enables you to seamlessly mix modes between autonomous and RC mode.
http://pdv.cs.tu-berlin.de/MARVIN/mark_ii_frameset_introduction.html
They have other cool tricks like calibrating angular sensors on an old record player’s turntable.

I don’t like the copters that much because of their noise and energy hunger and because my kids would cut their fingers (or more sensitve parts like the eyes) on some day. Observe that even MIT build an outer double bar rim protecting people that would stray on the corridors. Not that I have seen anyone near in the videos I’ve seen.
I would rather like to have a blimp like festo bionic mantas, but with more load capability and enough muscles to resist normal windy conditions… Still the load is not there, so I get tempted to process data off-vehicle again.)

Greetings!

Andi

p.s.:
Just because you process something in a base station doesn’t mean it is not acting autonomous. Or does it? When mobile computing power rises, one can later reinsert outsourced calculus, so I would reason.
Somethig different is MIT’s “aggeressive mode” Quad, where IR Cameras fix the bot to some specially designed room.
Still consider this: areas with GPS coverage are just the same “specially designed volumes” on earth, is a robot using gps non-autonomous?
Perhaps later on some swarm of differently designed robots could together explore an unknown volume and set up some presice space monitoring net, and distribute calculus needed amongst each other.

bye