Why autonomous tram comes long

“The staff is still fully responsible. Moreover, people do not dare,” explains Christian Zinner, senior research engineer in the Assistive & Autonomous Systems Area at the Austrian Institute for Technology (AIT). At the moment, the tram area is mainly concerned with avoiding collisions, so that passengers and road users remain intact and that very financial damage can be minimized. Even smaller incidents would activate traffic recordings if this leads to operational interruptions and trams are drawn up on trams.

To prevent this, the AIT has developed an assistance system that monitors the environment while driving and controls the speed. In particular, a 3D camera or other road users register in advance or will be. Potential collision risks can be derived from intervals and mutual speeds. If the risk exceeds a certain extent, the driver receives an acoustic warning. If this is not responded, the system actively introduces braking. “The goal is not necessarily to be better than people. The big difference is that these systems are permanently attentive,” said Zinter.

Speed ​​is also checked
In addition to avoiding collision, the driving speed can also be checked and compared with given speed profiles. “There have been spectacular accidents, because a black -out the tram drove too fast in the curve and then derailed or destroyed,” the expert explains.

There are no concrete figures about how strong collisions or incidents can be reduced by the two functions, “but if it didn’t bear fruit, the question would not be that strong”. According to Zinner, assistance systems are now almost standard for tenders for new vehicles.

Acceptance is fundamentally high among drivers, although there may be distortions during the feedback: “Nobody likes to make them a mistake. If the system slows down, this is often not reported.” It is important to clearly communicate what can be expected from the solution used and how it works. You might tempt too high perfection to trust too much and be less attentive. In any case, the technology is not delivered, you can “overlap” if detected dangers were not relevant.

No avoidance possible
From automated driving you are still a long way free, also because rail vehicles have a different driving dynamic than cars. The possibility of avoidance is missing, while the braking paths are much longer. That is why the sensor systems built on the tram must be able to look much further ahead. Solutions from the automobile area to transfer to trams therefore only produced moderate results. The maximum feasible inhibiting effect may also be used only in an emergency, otherwise the passengers – standing, not lying – can come to the fall. In general, the challenges in the traffic of inner cities are great.

The trams are equipped with stereo cameras in the solution developed by the AIT, which is used in Alstom (formerly Bombardier). These would guarantee a high recognition range and offer a very good side resolution. The latter prevents a train that comes to the opposite track and only passes by a few centimeters is seen as an obstacle.

Obstacle detection or compliance with the permitted driving speed are essential functions. “However, the requirements are higher for the automation of trains and the technology is not yet ripe enough,” said the researcher. The train without a driver in an open environment also differed considerably from structurally separated rail areas-keyword Metro. The much higher security level is the bottleneck. Nevertheless, automated driving is referred to in the railway area, where an operator can follow various trains at the control center without a train driver in the vehicle.

Company travel in the depot
With trams, the situation as a result of the complexity of city traffic is even more difficult. Here the automation in the depot is logical because of the many company travel – for service to the workshop, for cleaning or providing the line. But here too there are still many open questions. “The most likely you will be able to do this with completely new projects, so when a depot is rebuilt and a new vehicle fleet is purchased,” Zinter is convinced.

Artificial Intelligence (AI) is “still completely out of the game when driving from the point of view of security. It is currently not possible to bring such technologies approval,” says Zinter, referring to standards and regulations. Ki can be used very well for the functionality of assistance systems. However, the AIT model is still based on “solid physical models, real 3D sensors and actual measurements”. The situation is different with the real estate classification.

Recognize and allocate
A technology has been developed that is possible to map unknown routes. “The vehicle drives the rail network, where the network can be reconstructed fairly accurately from the collected data,” explains Zinter. Moreover, the area around the tram is interpreted semantically, ie analyzes whether it is a road surface, sidewalk, building, person or vegetation. Ideally, this Semantic 3D model succeeds in assigning each pixel of a certain object category. You know that from the automotive sector. “For us it was striking that there was nothing similar to the railway vehicles.”

From this work, the largest data set in the world for rail -related vehicles called “Railssem19” has emerged according to the information. Based on these training data, AI systems could be developed and tested for both the railway area and the tram environments.