ArduPilot's ModeReason enum is the exhaustive list of on-board autonomous decisions: of 56 values, 43 are the aircraft deciding for itself, 21 of those because it detected something dangerous, and exactly one — SOARING_THERMAL_DETECTED — because it found an opportunity. As for end-to-end, PX4 mainline's mc_nn_control has a 10 KB tensor arena and a Kconfig default of n; ArduPilot mainline has none.
Grepping all 9,199 lines of Skybrush's open-source light-show firmware for neighbour, collision and swarm turns up nothing: no aircraft knows another exists, and the entire content two hundred of them synchronise is one GPS time-of-week integer plus a millisecond offset, with collision avoidance settled on the ground in the choreography software. Chapter 7 of Taiwan's drone cybersecurity spec — the swarm chapter — leaves the 'drone' column a dash down all twelve items and tests switches and routers instead.
Flying a drone to 28 m in SITL and switching on the simulator's built-in GPS jamming, the flight controller declared EKF failure about seven seconds later, switched to LAND, landed and disarmed — it neither flew away nor crashed. Reading PX4's source then shows that of the 12 gates in EKF2_GPS_CHECK, bit 11, jamming detection, is off by default: the estimator catches jamming on its own, while spoofing can only be reported by the receiver.
After cloning and building both, three things differ from the stereotype: ArduPilot's EKF3 header credits the derivation to PX4/ecl, so the hardest layer is shared; PX4's last year of commits comes from company domains (380 from Auterion alone) while ArduPilot's comes from personal addresses with one contributor at 37%; and what really decides the choice is not performance but BSD-3 versus GPLv3, and which layer you need to modify.