From legacy to autonomous rail

We have found that introducing autonomous systems to a network of human-operated systems creates complex and difficult-to-predict behaviours. Also, performance benefits might not be seen until there is a high prevalence of autonomous systems across a network.

Sunlight shining into Empty railway station platform

PhD student: Emily Morey

Supervisory  team: K. Galvin and R. E. Wilson

A mix of technologies

Transforming large-scale networks such as transport infrastructure from legacy human-operated technology to using better performing autonomous systems is a major challenge. The required scale of investment means a prolonged time of older, human-operated devices and incoming autonomous technology operating alongside each other.

This mixing of technologies creates complex interactions and behaviours that mean the transitional system performance does not simply reflect the balance of the old and new approaches. Understanding this is important to manage and monitor the introduction of autonomous technology to a legacy network.

We have investigated this transition by modelling the safety limits and operation of a rail system that includes human-operated and autonomous goods trains.

Rail network characteristics

We found that different types of rail networks have different characteristics:

Human operation

Human-operated trains require long safety margins of track due to being limited by trackside signal control and limited driver information. This caused networks to stall very quickly above a certain track density of trains.

Autonomous trains

Autonomous trains potentially have far shorter safety margins than human-operated trains through information sharing across a network. We found that a fully autonomous train network could more than double the flow rate of trains seen with human-operated trains only.

Mixed operation

Operating human-controlled and autonomous trains in a network leads to highly complex behaviours. Low numbers of autonomous trains are usually surrounded by human-operated trains, which limits their performance, while the legacy trains offer poorer performance as they become blocked by the autonomous systems. This means there is little increase in network performance until a large proportion of trains are autonomous.

This provides insight for managing the transition of any network from legacy towards autonomous systems:

  • Mixing technologies results in behaviours that are complex and difficult to predict;
  • The benefits of introducing autonomous systems into legacy networks might only be seen once a significant proportion of systems are autonomous;
  • The performance of a mixed-technology network might be improved by managing where new, autonomous technology is deployed. For example, with rail, separate clustering of autonomous and human-controlled trains helps avoid the performance of each being reduced by the mixing of technologies;
  • Simulations are essential for proving the safety and potential capacity of any proposed technology changes to legacy-based networks.

The rich behaviour emerges despite having only two types of behaviour: fully human-controlled and fully autonomous. Real systems are far more diverse, with increasing levels of automation typically introduced over time. This greater diversity will only add to the complexity of operations.

Both the current and future work demonstrates the complexities that need to be navigated in a mixed-technology network before improved performance can be expected. Understanding this is critical to giving long-term confidence in investment decisions and in safety.