> the nascent sport
That is already 12 years old, so old enough that you can easily envisage someone who wasn't born at the time of the first races taking part. With 320 million households involved and at least half a dozen different organisations setting up races, aren't we well past "nascent", and have even missed the opportunity to call it a "fledgling"?
Anyway, more to the point:
> "Reaching the level of professional pilots with an autonomous drone is challenging because the robot needs to fly at its physical limits while estimating its speed and location in the circuit exclusively from onboard sensors."
The drone finds it "challenging" against a human when it has sensors to actually measure speed, location and, probably, also distance to obstacles! One a pre-determined course! They don't even have to worry about motion sickness from the FPV goggles (if you thought VR can make you puke...). What kind of wimpy algorithms are they using that it is even mildly surprising that the machine can't drive the course at the limits of the fligjt hardware's capability? Did all those maze-running mice die in vain?
> developed a system that combines deep reinforcement learning (RL)
Oh, they are using *that* sort of algorithm! No doubt they are more surprised it didn't suddenly flip upside down halfway around and run the whole course backwards (for no good reason that anyone can figure out).
Sorry, but given how many years have passed since clever chaps got drones to juggle sticks between each other, running a known course with sensors somehow doesn't seem terribly exciting.
Now, if they had run the course with just the feed from the FPV and the same level of control loop as the humans did, then we can talk turkey.
> could eventually lead to applications in battlefield warfare.
Just so long as everyone agrees on the course, eh? As Colonel Blimp said, "War starts at Midnight".
 if they are only estimating those things, go and buy some better sensors!
 if not, they really need to read some more hobbyist electronics blogs!