A technical overview on the use of Artificial Intelligence (AI)/Machine Learning (ML) illustrates how it can be applied in autonomous passenger and freight trains.

It describes how, on train wagons, AI/ML applications can be connected to a single Gigabit Ethernet network that uses the Common Industrial Protocol (CIP). The network connects all sensors, controllers, and actuators in the wagons, as well as wired and wireless entertainment, and feeds the information to the AI/ML applications.

Artificial Intelligence/Machine Learning (AI/ML) subjects (reasoning with uncertainty, autonomous path planning, collision avoidance, swarm technology, sensor fusion, etc.) are used to support the use cases of trains and trolleys.

Grade of Automation

The IEC 62290-1:2014 standard defines various levels of automation and the minimum mandatory functions required to achieve these levels referred to as Grades of Automation (GoA). This standard focuses on rail automation for public transit systems such as subways, metros or commuter trains referred to as Urban Guided Transport (UGT). The standard defines five levels of automation:

• GoA 0 – Line of Sight Operations
• GoA 1 – Non-Automated Train Operation
• GoA 2 – Semi Automated Train Operation
• GoA 3 – Driverless Train Operation
• GoA 4 – Unattended Train Operation

The IEC 62290-1:2014 standard defines a set of minimum mandatory functions a system must implement. These mandatory functions are:

• Ensure the safe movement of trains: Ensure a safe train route; Ensure a safe separation of trains; Ensure a safe speed.
• Drive the train: Control acceleration and braking.
• Supervise guideway (i.e. train tracks): Prevent collision with obstacles; Prevent collision with persons on track.
• Supervise passenger transfer: Control passenger doors; Prevent injuries to persons between cars; Prevent injuries to persons between platform and trains; Ensure safe starting conditions.
• Operate a train: Put in or take out of operation; Supervise status of the train.
• Ensure detection and management of emergency situations: Detect fire/smoke; Detect derailment; Detect loss of train integrity; Manage passenger requests (call/evacuation, supervision)

The mandatory functions are either the operation staff responsibility or performed by an automated/autonomous system.

GoA 0 – Line of Sight Operations

In the GoA 0 level all train movements and control of wayside elements (such as track switches) are managed by manual operational procedures executed by an Operation Staff. There are no mandatory functions executed by an automated/autonomous system.

For example, if a train is required to move from a storage lane to a transfer track, it is the responsibility of the train operator to ensure the switches are in the correct position and ensure the train does not collide with the train ahead.

For example, if a train is required to move from a storage lane to a transfer track, it is the responsibility of the train operator to ensure the switches are in the correct position and ensure the train does not collide with the train ahead.

GoA 1 – Non-Automated Train Operation

A GoA 1 level system is considered a conventionally signaled system common to all subway or metro systems. The IEC 62290-1 specification has allocated the “Ensure Safe Movement of Trains” mandatory functions to the Automated/Autonomous System.

The Automated/Autonomous System uses Automatic Train Protection (ATP) will: determine if it is safe for a train to proceed by locking switches and setting the aspect of the signals, checks that speed is compatible with permitted limits and activate an emergency brake if necessary.

The Operator/Driver is responsible for: opening and closing train doors and determining if it is safe to depart a platform, control braking, propulsion and reading the wayside signals accurately before moving the train forward, responding to emergencies or sudden changes.

GoA 2 – Semi Automated Train Operation

The GoA 2 level is where an automation/autonomous system becomes more effective. This is one of the most-common levels of automation today. The jump from GoA1 to GoA2 is an order of magnitude higher than the jump to any other grade (for example from GoA2 to 3 or GoA3 to 4) due to the complexity and amount of automation required.

The Automated/Autonomous system is responsible for the following processes: accelerating and stopping the train, supervising speed, and braking when exceeding speed limits, determining the train’s position on the track, transmitting data from the vehicle to a wayside monitoring system and requesting authorization for the route.

The Operator/Driver is still present to perform several vital tasks as well as take over if the system should fail: Stopping the train if there are obstructions or if a person has fallen on the track; Be mindful of crews working alongside tracks and appropriately adjust speed; Open and close train doors; Authorize departure from the station.

GoA 3 – Driverless Train Operation

In GoA 3 level, the automated/autonomous system controls the train between platforms. There’s no driver, but there are attendants on the train. In this level, starting and stopping are automated and a train attendant operates the doors and drives the train in case of emergencies.

A train equipped with a GoA 3 automated/autonomous system is aware of its surroundings along the track. Trackside devices installed along the track detect obstacles that may obstruct train movement such as tunnel ventilation doors or work crews performing maintenance at track level.

These devices connect to a wayside unit that sets the movement authority for all trains.

All core operations of GoA 2 are covered and the system additionally: Monitors all platform elements; Stops the train if there are obstructions which are partially or fully on track; Protects crews at a track level.

Attendants on board are assigned to take care of the following tasks: Open and close doors; Authorize departure from the station.

GoA 4 – Unattended Train Operation

The GoA 4 level is where starting and stopping, operation of doors and handling of emergencies are fully automated without any on-train staff. The train’s health will be monitored remotely.

At this level, the system is in complete control over the train, wayside and the platforms. Involvement from the operator has been reduced if not eliminated. The operator’s role is to only monitor the system and get involved if there is a failure the automated system cannot handle.

The train or tram is in full control of all systems on track as well as in the station: Opens and closes doors; Determines if it is safe to depart from a station; Performs self-tests; Detects emergency situations; Can take the train in and out of the depot.

GoA 4 has also extended its reach into the yard which means human intervention is no longer required to bring a train from the storage tracks to the mainline. GoA 4 will “wakeup” the train, perform self-tests, launch the train into the mainline, route the train to the yard at the end of service, park the train in the storage track, command the train to “sleep” and repeat the process the next day.

Autonomous Train Control System

Figure 1: Autonomous Train Control System.

The Autonomous Train Control System (ATCS) in Figure 1 describes the Automatic/Autonomous system that will support the Grades of Automation.
The ATCS defined here consists of three types of autonomous subsystem types. These subsystem types are:

• the On-Board Control System (OBCS),
• the Wayside Control System (WCS),
• and the Operation Center. The Operation Center is not discussed in this document. All subsystems consist of the same System communication features.

On-Board Control System

The On-Board Control System (OBCS) identified in figure 1 consists of Autonomous functions (illustrated in Figure 2) that may use Artificial Intelligence /Machine Learning (AI/ML) algorithms to accomplish several tasks.

Automatic Train Protection (ATP) is the autonomous function that performs the following tasks to protect the train:

• Train Localization: scans train surroundings to determine anomalies.
• Train Speed Monitoring: Monitors the train speed and compares to path projection speeds.
• Movement Authority: monitors environment for anything that may hinder the progress of the train.
• Track Resource Control: determines wayside resources around the train required to operate, track availability, to allow the train to move forward.
• Static/Dynamic Speed Profiles: Maintains the speed profiles for the path of train.
• Braking Intervention Curves: Continuously calculate braking distance requirements based on current train speed and known terrain parameters.

Automatic Train Operation (ATO) is the autonomous function that performs the following tasks to operate the train:

• Route Monitoring: monitor the route (tracks and environment) to ensure there are no impediments to train progress.
• Platform Monitoring: Monitors the safe approach and capabilities of a platform the train is going to stop at.
• Precise Stop Control: control the braking of the train to make precise stops.
• Train Speed Control: Control speed.
• Schedule Management: determine what actions are needed to maintain the train schedule.
• Door Control: Control opening and closing of the train doors on passenger trains.
• Intelligent Driving Subsystem: to be discussed later.

Safety Protection (SP) is the autonomous function that performs the following tasks to safely operate the train:

• Lateral Intrusion Protection.
• Collision Protection.
• System Integrity Protection.
• CIP Safety.

Perception and Localization (PL) is the autonomous function that performs the following tasks to assist in safely operating the train:

• Near-field Surrounding Perception: an AI/ML algorithm that processes images of the surroundings of the train to determine safe passage.
• Image-LIDAR Fusion Perception: AI sensor fusion algorithm that combines sensor data to determine safe passage.
• SLAM-IMU-GNSS Localization: Navigation algorithms to determine the location of the trains and the trains proximity to other trains and objects.
• Virtual Track Line Perception: AI/ML algorithm for comparing the track in the train path for any obstructions.

Intelligent Driving Subsystem and On-Board Control Interfacing

The Autonomous Train Control System in Figure 3 describes both the Intelligent Driving Subsystem and On-Board Control Interfacing features.
Intelligent Driving Subsystem (IDS) is the autonomous function that uses AI/ML algorithms to perform the following tasks to drive the train:

• Automatic Path Tracking.
• Autonomous Guided Trajectory Following.
• The AI/ML algorithms will use Virtual Rail Forms that consist of: Painted Graphics, Magnetic Nails and Stripes, HD Maps, a Track Database (DBA) and Train information.

On-board Control Interfacing (OBCI) is the autonomous function that Interfaces to the train sensor and control equipment (i.e. Transponder Reader, Tachometer, Brake).

Figure 2: OBCS Autonomous Functions.

Figure 3: IDS and OBCI Diagrams.

Wayside Control System

The Wayside Control System (WCS) identified in figure 4 consists of Autonomous functions that may use Artificial Intelligence /Machine Learning (AI/ML) algorithms to accomplish several tasks.

The On-Board Control System (OBCS) identified in figure 1 consists of Autonomous functions (illustrated in Figure 2) that may use Artificial Intelligence /Machine Learning (AI/ML) algorithms to accomplish several tasks.

Figure 4: Wayside Control System.

The Autonomous Train Control System in Figure 4 describes the Automatic/Autonomous system that will support the Grades of Automation. Automatic Train Supervision (ATS) is the autonomous function that performs the following tasks to monitor and supervise the train:

• Train Monitoring.
• Operation Management.
• Track Resource Monitoring.
• Schedule/Route Monitoring.

Object Controller (OC) is the autonomous function that controls various machines supporting the train:

• Point Machine Control (Track Switches).
• Platform Screen Door Control.

Use Cases

Assisted (Train) Operation helps drivers avoid collisions with obstacles on the rail line. Assists the driver in monotonous or difficult traffic situations to minimize consequences and costs of accidents. Assisted Operation provides increased driver and passenger safety along with higher availability of trains and tracks. The benefits are:

• Avoid or reduce damage from accidents: Lower Repair costs and Higher train availability
• Improved punctuality and flexibility,
• Smooth acceleration and braking provide improved passage comfort,
• Improved safety.

Driverless Depot Operation institutes autonomous driving in the depot on semi-protected track areas at low speed. The benefits are:

• Operational even with reduction of personnel. Example: Reduced staff due to COVID,
• Reduction of accidents involving people in hazardous zones,
• Optimized vehicle transfer: Shorter wait times.

Driverless Vehicle Stabling and Provision is the automated transfer of trains to the stabling yard at end of operations as well as the automated provision of trains at start of operations. The benefits are:

• Reduce non-productive, time-consuming vehicle transfers,
• Faster provision of additional rail vehicles at time of peak demand,
• Trains can operate in parallel instead of waiting for worker to hop between trains.

Driverless operation on Specific Sections is the Autonomous System taking over the complete driving responsibility in suitable areas like fully segregated line sections. The benefits are:

• Fewer non-productive, time-consuming vehicle transfers,
• Drivers can be deployed elsewhere in passenger rail operations,
• Improved safety for all users.

Fully Automated Operation on Entire Lines is the driverless and unattended train operation on entire lines. The benefits are:

• No need for drivers or conductors,
• More space for passengers by elimination of driver’s cab,
• New operating models possible: shorter headways, demand-driven train operation and elimination of fixed timetables.

Additional autonomous benefits

Increased capacity: Automation expands passenger capacity by increasing the number of trains a line can handle. Many networks are already operating at full capacity and simply cannot handle additional trains, but automation creates space by eliminating the variables caused by human behavior.

Greater flexibility: Automating train fleets creates predictability, and that predictability enables greater flexibility. When supported with Automatic Train Operation, operators have more choices. They can automate service on seldom used or remote lines, automatically park a train, quickly respond to changes without needing a human driver to always be present.

Lower costs: Services like maintenance account for a large part of the costs of operating a fleet. Autonomous train technology ensures that trains are operated as efficiently as possible; reducing wear and tear while minimizing the incremental inefficiencies that are natural to human influence.

Enhanced passenger experience: Automatic train operation is not just for engineers and operators, it greatly improves the travel experience for passengers. Supported by technology, drivers can focus on passenger attendance and strategic decisions, meaning passengers will notice smoother acceleration and stopping, more comfortable curves and seamless transport options that have fewer delays.

More sustainable: Eco-driving rail automation systems reduce energy consumption and are key to making the rail industry more sustainable. Automatic train operation orchestrates rail fleets, empowering operators to run their trains more efficiently. Performing as a single connected system, this improved synchronization translates into maximum performance that helps reduce emissions and limit energy use.

George A. Ditzel, Industrial Communication Network Architect, Schneider Electric