AI and Communications, Freight Rail, Rail industry news (Australia, New Zealand)

AI for rail transport operations

4AI Systems is looking to bring a highly targeted AI system to the rail industry to support a safer sector.

4Tel is the parent company for 4AI Systems and began down the path of artificial intelligence (AI) in rail almost a decade ago. It soon became apparent that the AI division could support a separate business.

Chief technology officer Mark Wood spoke at the Heavy Haul Rail conference, providing an insight into AI and what it could mean for the rail industry.

“4Tel actually started down the AI path in 2016 and we formed 4AI Systems to focus in that area and allow 4Tel to remain in the rail space,” he said.

By focussing on the AI technology, the company has quickly developed an impressive system designed to support the rail industry.

4AI Systems deploys customised artificial intelligence and machine learning systems for transport operations.

The company was founded in 2021 as a subsidiary company of 4Tel, drawing on 24 years of experience in delivering industry-leading solutions to the rail industry. The specialised team comprise of professional AI scientists, mathematicians and engineers.

Separating the focus on artificial intelligence from 4Tel ensured that 4AI Systems could invest and develop emerging AI technology at a much greater scale to support the rail industry.

“We see the business case now as helping both the onboard operations and the offboard maintenance and the information provided in that operational environment,” Wood said.

AI for rail

Using sensors from the driverless car industry, the 4AI Systems team has developed ‘HORUS’, an AI system to provide greater situational awareness to train drivers performing operations in the rail corridor.

4AI chief technology officer Mark Wood spoke at the Heavy Haul Rail conference in Perth. IMAGE: 4AI

”Rail corridors rarely change, making it ideal to explore and develop real-time visual detection software. 

By capturing the corridor master sequence and then using an on-board supercomputer to perform real-time visual processing, the corridor information is cross checked with a detailed database enabling the system to detect if a trespasser or object is in a potentially dangerous position in relation to the train’s path.

While the system won’t stop incidents from happening, it will minimise the impact by eliciting an earlier response.

Wood explained how the organisation has developed the system over the years.

“We have a good collaboration with University of Newcastle and we draw on their robotics and AI areas to supplement our own research and development,” he said.

“What is really interesting is that with our approach, we understand the weaknesses that some sensors may have, and then supplement that by other sensors.”

This development has been important for 4AI Systems to develop what it believes is a complete solution.

How it works

Each locomotive or railcar needs to be fitted with any combination of advanced sensors to gather information, including GPS, inertial navigation, cameras, and optionally, radar, lidar, and ultrasound sensors, with other incidental inputs such as odometers, speed and temperature sensors.

All of this data is then collated to support the AI system. Fusing multiple-sources of sensor data frame-by-frame and integrating it over time, the locomotive or train inference computer can use AI to support to support driver decision making and situational awareness.

Mark Wood explained the mission of 4AI Systems in greater detail.

“Our mission is to provide solutions that empower people and organisations to ultimately make better, data-driven decisions in their rail operations. We believe AI is the next step in creating more efficient rail networks,” he said.

“Onboard AI minimises investment into third-party infrastructure systems because it creates smart trains owned by the train operator. AI handles boredom better than humans. AI never gets tired or distracted, has better visibility, can monitor location, speed and route in real-time, and assess track infrastructure for risks.

“You can highlight a person as they walk around the corridor – they are highlighted red when in the rail corridor, yellow in the warning area and green when they are at a safe distance.”

4AI Systems has been developing systems to deliver the right kind of information to the driver to not overwhelm them but also provide what is necessary.

“We need to ensure we do not distract the driver and there are a number of ways we can provide that information to them,”
Wood said.

“A warning will appear in the same spot on the screens so the driver knows where to look when an alert is heard.

“A simple graph can show where an obstacle may be as opposed to live pictures.

“A driver may want additional information, or access to the cameras in high-risk scenarios, and they can gain access to that.”


The HORUS system can provide accurate location definitions to a specific track and location in multiple-track territory by fusing imagery and navigational data.

It can detect an unexpected object or hazard that may be on or around the rail track.

It also provides speed checking and authority enforcement. 4AI Systems can recognise signal posts and reading of displayed aspect. It can also monitor the condition of infrastructure to report back to maintenance teams and provide real-time feedback.

“This type of real-time data can minimise downtime for maintenance on tracks and even minimise the instances of accidents,” Wood said.

The HORUS system can work in low light as it allows a range of sensors to work together to provide a full picture. IMAGE:4AI

The system can also track events, processing all log files, cameras, accelerometers and speed data onto an event sequence file that can then be replayed.

Another scenario HORUS can assist with is adding situational awareness around level crossings improving level crossing safety.The HORUS system allows for warnings and alerts for approaching and around high-risk areas.

Algorithms can calculate probable hazards and provide a warning classification to ensure an appropriate response to potential hazards.

 “When it comes to level crossing detection, a few seconds extra can make all the difference,” Wood said.

“There is an ability to visualise vehicles and people that are travelling towards the danger zone and we can alert the driver ensuring that they get an early warning.

“A few seconds can make the difference between a fatality and a near miss.

“You can’t beat the laws of physics. If you can buy a few seconds, it might be the difference between someone going home safely at the end of the night or not.”

Real world success

In 2022, 4AI Systems completed an AI vision system feasibility study for Rio Tinto Iron Ore’s AutoHaul in Western Australia’s Pilbara region.

Rio Tinto Iron Ore installed a 4AI Systems HORUS AI solution on a locomotive to learn and understand the system’s capability and potential as a complimentary technology solution.

Five cameras and other sensors were installed on the AutoHaul locomotive to great success.

The program demonstrated how AI software can detect hazards in different conditions and from different distances.

The study was able to identify people and objects beyond 1000 metres including around curves with line of sight. The system performed continuously well in all weather conditions. 

The system has also been trialled on New York’s metro network. This trial gave the team a unique insight into the New York subway operating conditions.

“Our team was very excited to test the system again in New York’s colder weather conditions. The data provided was an interesting contrast to Pilbara’s dry heat,” Wood said.

It was exciting for us to see our system working across a range of environments and in turn being successful across those regions.”