As far as I am aware, there has been no non-supervised train in a general train network. (They exist in closed systems.) However whether the train is self-driving is really not the issue here. In current railway safety systems (even from the 80s) a crash is about the same probability as random bitflips. However reality is too messy, so that the availability is not satisfactory enough to not use a human fallback, and then you are back to the assisted human error rate.

Current operation of a train is basically self-driving and supervising driver. So if you suggest "just use self-driving", you make the system less safe, not more, because it means just getting rid of the human supervisor. This is not currently done, because it would be too insecure. It might be done in the future, but rather because less safety is going to be accepted in favor for getting rid of the costly human.

A self-driving train will happily cause a crash, if the target is "drive with high-speed in that segment currently occupied by another train". The issue is that that becomes a target, not that there is a human in the loop.

> A self-driving train will happily cause a crash, if the target is "drive with high-speed in that segment currently occupied by another train".

Sure, if you allow trains to ignore safety completely, then accidents will occur. That is a ridiculous statement.