Railway signalling IMO is an area where a little research could dramatically increase the throughput of the railway network with a rather low cost.

Todays railways mostly use fixed block signalling. Expensive and unreliable equipment ensures that there is only one train on each 'block' of railway track at the same time. That forces trains to be at least 1 or 2 blocks apart, which are frequently multiple miles long. End result: Trains are usually 10's of minutes apart, or at least 1 minute apart even in urban subway systems.

A more advanced system would aim to have trains as close together as cars are on a freeway. Trains would be able to hitch and unhitch whilst moving 100 mph, allowing different loads to be sent different directions and different sets of passengers to stop at different stations. Crashes would be avoided by having every train know about the train ahead and behind, and unable to make any move which would cause a collision (ie. it is not allowed to slam the brakes on if there is a train right behind you).

Such a system would be implemented primarily on the trains, and primarily in software. Software is expensive to develop then cheap to replicate, a property very important if you want to deploy it widely.

It would be backwards compatible with fixed block signalling by simply saying that within each existing fixed block there is either one legacy train, or an unlimited number of gen 2 trains who will all communicate with eachother (and any train who cannot communicate with every other train within the block is not allowed to enter the block).

Hardware/software failures are kept safe by simulating every possible type of failure (ie. braking, locomotion, power, derail, position uncertainty, comms), and every second generating a plan which will be executed by every train in case of each type of failure occurring. Any move any train wants to make where any of the above plans cannot be generated with a safe outcome isn't allowed.

SimSig: Railway Signalling Simulations