Note that both the Arduino, ESC and RF modules use the same ground, (I think that's the problem).
Yeah, quite possibly... or more to the point, the problem is there is no ground, it's all spaghetti.
Wires are not short-circuits at radio frequencies. What you think of as one net, at one point in the circuit, in general, does not have the same voltage at some other point. That wire has traveled through a varying electromagnetic field, and has some voltage induced on it in addition to the source signal. And similarly for all ground connections. "Ground" itself is meaningless in such an environment: we can consider local ground at a circuit, or component (say, an IC's VDD/GND pins, and signals with respect to that), but in the space between, where wiring flows, both connections pick up indeterminate induced voltages (or currents) and the whole thing is a nightmare to troubleshoot.
It's very likely the ESC switches its outputs very fast (say, 10s of ns), which launches a wave at the motor, and an inverse wave up the supply and control pins. As the wave induces into nearby wiring, logic levels are momentarily corrupted, and sensitive circuits can be disturbed. In more extreme cases (like ESD), clamp diodes are activated and stray currents can flow all over the place (including triggering CMOS latchup). I think that extreme is unlikely, but it could also be that some parts are in a really unlucky position and things have conspired to have a similar effect.
The Arduino doesn't seem to have too much trouble with that (if you were reading any inputs from the ESCs, they would probably need filtering or error correction), but the comms between Arduino and nRF, or the nRF itself (noisy supply?), aren't going to be doing well.
Note that this isn't, or isn't necessarily, corruption of the radio itself -- I mean direct interference with the antenna and signals. The ESCs probably don't produce significant if any harmonics that high up (whatever it is, 433MHz, 2450MHz, etc.). More likely routes of interference are poor supply voltage quality (more and better filtering), corrupted signals (induced and ground-loop), or corrupted analog functions like clock oscillator or PLL, etc.
The first line of defense is to use a structure that permits some manner of analytical sanity. If we can always have a nearby point to measure voltage against, we might still have induced voltage between points on that surface -- but at least we can read a signal consistently. This surface is called a ground plane.
Try adding a sheet of metal (copper or tin plated steel is effective for this) underneath all the circuits, and attach all ground connections to it. More pins, the merrier -- again, wire length counts against you, keep grounds short and many. Keep wires low, close to the ground plane, to greatly reduce induction between them. Route the ESC power lines some distance away from everything else -- they should only be common where the battery and regulator connect. The ESC power lines will still have to cross signal lines by the looks of it, but this can be done with a brief hop across the signals, minimizing coupling. A strip of grounded metal could also be placed between the wires to further shield them, but this probably won't be required (you also may not have any way to determine if such a minor change is actually the problem).
The ESCs draw considerable currents, and quickly (albeit on ~ms time scales, not the ~ns we're talking for EMI here). This may still drop relatively a lot of voltage along the plane, due to its resistance (particularly if steel). You should probably keep their ground wires, and just strip some insulation near the unit to make a local ground connection. That way the ground wires can be RF-grounded where important, but the wire is still the dominant (LF/DC) current flow path.
