re:
http://newatlas.com/robot-alias-darpa-cessna-caravan/45973/At first I thought that was dumbest thing ever. Then I thought about it a bit and could see the application for retrofitting aircraft. In many (most/nearly all..?) existing aircraft there is no common computer box that connects
everything on-board. The only place
something could control the fuel pumps, cabin pressurization, radios, and flight path is in the cockpit using the physical switches. Changing out computer boxes is very hard, adding software to existing flight systems is very hard because of regulations, limited computing power, vendors no longer supporting devices and shit just being really old. Yes many aircraft already have a software box that can fly way points but almost certainly the only 'API' is physical buttons.
The other hassle is that even for a given type of aircraft there can be different manufactures under the hood who made the autopilot and flight systems or they were updated at different times. So you may have 20 Boeing 767 you want to turn into UAV's but some run Windows XP, some run Win 95, some run Win 10, some run Mac OS8 etc. They probably all function 98% identically at the pilot level but can be radically different under the hood where if you wanted to add some new remote pilot system you would have to make +6 unique systems. The economics here get really stupid to make one-off custom systems to interface directly into each 'OS'; or you could make a common "robo-pilot" like in the linked article that takes advantage of where they all already are nearly identical.
Flying a plane is much more than moving the yoke around, if you want to turn a plane into a remotely piloted aircraft (or 100% automated for that matter) you need to be able to affect all on board systems.
"why not just gut all the electronics and put in new fancy ones that talk to everything": That would require re-certification with the FAA and that is VERY expensive. Also your system would still need to talk with the 10 different engine control computers (among others) that are in the fleet so you would still be doing custom work on 10/20/30 year old computers/software.
Is the "robo-pilot" a good idea when building a new UAS from the ground up? No. Could it be a good idea for retrofitting existing aircraft? Maybe.
When the MythBusters need to make a car remotely controlled they dont go into the onboard computers, they put actuators on the wheel/brake/gas, this is the same thing.