They need to talk to their friends over at FHWA and TRB. They've been studying automated vehicles, sensor systems, and communication for years. Unfortunately, they haven't been able to commercialize much of it, and now FCC has sold the radio spectrum they needed for V2V and V2X out from under them (Verizon needs to make more money!). Instead, we have Tesla making all of its customers guinea pigs with "self-driving" that sort of works (pay attention!!!!!); GM where it only works where things have been very carefully mapped, at least for the very rich public (see: SuperCruise in Cadillacs); and GM and Google in carefully chosen environments at low speeds.

Railroads should be able to do better. At least the robot wouldn't need to steer, and if it can't stop in time, well, the human wouldn't be able to do much better (what computer sensor is needed to feel the small bump when the locomotive crushes a car or a trespasser?). I'd see a system more like GM's (carefully mapped, courtesy of PTC) than Tesla's (perpetually beta, and trying to see without Lidar). What I don't see, is an effective system coming out of basic government research into sensor tech at this point. The sensors are a known deal. The need (if you want automated trains) is to integrate them into a working system before Congress mandates one on an impossible schedule, and to fence all of the r/w (with razor wire and electric fences) and eliminate all the grade crossings.