Start-up of the Day: Vialytics quickly registers road conditions

How do self-driving cars handle potholes on the road? As just stay driving ahead or spontaneously around them aren’t an option. You have to take the bull by the horns, that’s what the founders of vialytics were thinking. They designed a system that uses artificial intelligence (AI) to map out road conditions. This is how the road authorities can deal with the problems as quickly as possible. Danilo Jovicic, who founded the start-up together with Achim Hoth and Patrick Glaser, explains how the system works.

The founders of vialytics GmbH, (from left to right) Achim Hoth, Patrick Glaser, Danilo Jovicic ©vialytics

How did you come up with the idea of setting up vialytics?

We wanted to do business as an independent company and set up our own start-up. We got to know each other through the Activatr and Pioniergeist start-up programs. It was by coincidence that we then got together in a small group. That’s where the idea of doing something with road management took shape. We came up with a lot of wild ideas for a couple of weeks. We also had a lot of contact with municipalities who told us about problems concerning road management. The overarching issue there was autonomous traffic. We thought carefully about what you need to do in order to be able to drive safely autonomously. That invariably comes down to good roads.

What does your product look like?

Municipalities can continuously monitor their streets with our system. This is done with the help of a modified smartphone mounted on the windscreen of a municipal service vehicle. On a sweeper, for instance. These are at any rate always out and about in the city. The smartphone records the road every 4 meters.

This data is subsequently sent to us. It is then analyzed using an algorithm. Any damage to the road is automatically detected this way. The municipalities get the data back again in the form of a dynamic map. As they are better informed about the condition of the roads, they can react more quickly to any damage. This leads to a more sustainable and efficient way of road management. After all, plenty of municipalities don’t address the maintenance of their streets until it is far too late. Which means that the costs are also much higher. Current systems do not offer a proper solution. Those recordings are actually made with too great a time frame between each other. Nor are they carried out systematically.

Was there a problem you had to resolve first?

It was particularly difficult in the beginning to gain the trust of municipalities. This was mainly due to the fact that municipalities rarely cooperate with start-ups here. We set up 5 pilot projects where our system was tested. Thanks to the positive reactions we received, we have now managed to build up a customer base of 50 municipalities throughout Germany. Currently, we are also in contact with cities in other countries who are interested in our product.

What are you especially proud of?

We are especially proud of our first customers who have dispelled any preconceptions that local councils are a bit stuffy. Some of them were so enthusiastic about our solution that they bought the system before it had even been fully developed. Of course, we are also very proud of our team, which has expanded considerably over the last 6 months. Our employees are busy developing the product on a daily basis.


What does the future of vialytics look like?

Our goal is that of internationalization. We want road authorities all over the world to be able to maintain their road networks in an efficient and sustainable manner. Apart from that, we will continue to work on improving things so that we can keep on responding to the requests of our customers.

What tips do you have for other starters?

Do you have a good idea? Jump into the deep end and dare to make your dreams come true. And for those who have already set up a company: at some stage, take each employee along with you to a client. That’s what you’ll learn the most from.

More articles on start-ups can be found here.


Autonomous cars: Overtaking is only possible with a human driver at the wheel

So far, autonomous cars are only dreams of the future. Scientists around the world are working flat out to find out how a car can navigate everyday traffic without human intervention. Automated vehicles must not only be able to navigate correctly and recognize obstacles, but must also assert themselves over human drivers. For example, they need to be able to overtake slower vehicles (IO has previously reported on this).

However, not just fully autonomous cars have to overcome this hurdle, but also the highly automated cars that are currently already on the road. Overtaking maneuvers on two-lane country roads pose a problem in particular for built-in technology. That’s because the vehicles out in front interfere with the sensors. In order to solve this problem, scientists at the University of Ulm in Germany have now devised a system that is able to perfectly divide tasks that overtaking entails between humans and the system.

More articles on the subject of autonomous driving can be found here.

Blind Sensors

Passing a vehicle on two-lane roads is often not easy even for a human being behind the wheel. Especially if you have to assess whether it is safe to overtake a vehicle on the other side of the road. Or if the vehicle in front of you is a truck and is blocking your view. The sensors of an autonomous or highly automated car are already disrupted by a normal passenger car. So, if bends, uphill or downhill slopes are added, it becomes even more difficult for the computer system to get an overview of the situation.

“People know that overtaking in such situations is highly risky,” explains Marcel Walch, principal author of the winning Best Paper Award publication. Together with computer scientist and university president Professor Michael Weber, and psychologists from Ulm University, the doctoral student from the Institute of Media Informatics has designed a cooperative system that combines the respective strengths of the driver and vehicle.

© Marcel Walch

While people are better able to grasp a traffic situation in spatial terms and assess the associated dangers more ‘realistically’, the technology is impressive as far as road handing and maneuverability is concerned, explains Walch. It is therefore only logical to combine the strengths of humans and technology when it comes to overtaking actions. Otherwise, the vehicle would have to drive even further behind the slower vehicle that is in front. Or else the driver would have to overtake ” manually. ” In other words, the human driver has to make the decision to overtake, but the maneuver itself is carried out by the vehicle. For this transition a “handover” is necessary, which enables a switch from automatic to manual operation.

When the other lane is clear, people assume that the automation system will initiate an overtaking maneuver. Professor Martin Baumann, head of the Human Factors Department (who is participating in this research project along with two doctoral students) emphasized that trust in technology can be compromised and its acceptance adversely affected if people experience automated behavior as inappropriate. However, if the technology indicates that it is not overtaking because its line of vision is limited, the human driver knows that they should intervene. They can tell the technology that the oncoming lane is clear. This way, humans can better “understand” the technology and provide adequate support, say the researchers. Driver and automation work together perfectly then.

Preliminary tests successfully completed

This cooperative system for overtaking was tested in a state-of-the-art driving simulator at the Institute for Human Factors. A large display has been integrated into the simulator’s cockpit which enables the driver and vehicle to interact. The terrain and the travel route on a two-lane country road were projected onto three large screens. The speed was around 100 km/h in the simulation and the vehicle was required to overtake slower-moving cars which were driving on a varying route at around 70 km/h.

During the simulation, the human drivers had to handle certain test tasks which were designed to distract them to a greater or lesser extent. Since the vehicle was programmed to drive close behind the slower car in front so that “visibility” was restricted both for the driver and for the vehicle’s sensors, “dicey” situations often occurred. As intended and as often happens in real life. Then the overtaking process was either stopped automatically or could be stopped by the human driver.

© Marcel Walch

Press the button to overtake

Researchers have also explored what a viable interactive system could look like that would allow the driver to initiate overtaking. And – in case of oncoming traffic – stop it safely and on time. They examined two different intervention techniques: The “CLICK” and the “HOLD” procedure. With the “CLICK” method, the simulation participants had to press an “Allow overtaking” button on a display to overtake. This then became the “Cancel overtaking” button. This means that the overtaking process is aborted if the driver touches that button a second time. With the “HOLD” variant, the driver must keep the overtaking button pressed down until the change to the other lane has been completed. In the end, a lot of testers stated that they considered the “HOLD” technology to be safer. This was because the overtaking process could be stopped more quickly in the event of danger. Yet they also considered the “CLICK” variant to be more practical and user-friendly.

When the study was completed, it became clear that many test drivers preferred a cooperative approach to overtaking rather than just a manual one. At the same time, however, it was also clear that people are not always able to cope with complex situations. Especially when they are very distracted.  People often forgot to look in the rear-view mirror when an overtaking maneuver was initiated. The researchers therefore recommend that the vehicle should remind people to look in the rear-view mirror. And, if necessary, warn them of upcoming rear traffic by using sensor data in order to prevent potentially dangerous situations.

Cooperative driver-vehicle interaction

The project is part of the KoFFI joint project on cooperative driver-vehicle interaction. Which is funded by the German Federal Ministry of Education and Research. The project was presented at the Automotive UI 2019 in Utrecht and awarded a prize there in autumn this year. This is the leading trade fair for automotive user interfaces.

A similar project has also been carried out at the Karlsruhe Institute of Technology (KIT), where the PAKoS project will be presented at a driving demonstration on the 15th of November. It is also concerned with a smooth interaction between humans and technology and a problem-free situation-based handover to the human driver. Examples of these types of situations are construction sites. Places where there are speed limits and no clear lanes for the system, or non-mapped private areas.

Self-driving cars will never be possible in Amsterdam city center

Anyone who thinks that the self-driving car is the future is wrong. At least when it comes to the chaotic mess in city centers like Amsterdam. These are so cluttered and unpredictable that it would be impossible for autonomous vehicles to anticipate traffic conditions. Which is what Carlo van de Weijer has predicted at the opening of the AI in Engineering symposium. He is director  of the new Eindhoven Artificial Intelligence Systems Institute (EAISI, pronounced ‘easy’ in English) at TU Eindhoven in the Netherlands.

Traffic chaos in the Amsterdam city center is too much for self-driving cars. Photo: Lucette Mascini

Experiment in the US

Van de Weijer worked in the automobile industry for a long time and wanted to answer the question as to why we are still not driving through the country in self-driving cars. The reason is that it is very difficult to make automated vehicles function like a human being. Van de Weijer gave an example of an experiment in the US where a robotic car kept driving on the right lane while a very slow truck was driving on it’s left. A tailgater behind the autonomous car wanted to pass but wasn’t able to.

Read other Innovation Origins columns by Carlo van de Weijer here.

Read moreSelf-driving cars will never be possible in Amsterdam city center

Waymo’s autonomous cars are allowed to drive in public

The Californian licensing authority CPUC has granted the Alphabet-daughter company permission to have vehicles controlled by the company’s own system driven on public roads. This makes Waymo the fourth supplier to receive such a permit, alongside Zoox, Autox and By the way, was recently taken over by Apple.

The news fittingly comes around the same time as the recently announced cooperation with Renault-Nissan-Mitsubishi, one of the largest car manufacturers in the world.


The permit comes with strings attached. The Waymo vehicles are not allowed on the road without a supervisor on board. In addition, it is not possible to use the vehicles in a taxi or Uber manner. Instead, Waymo employees and guests may be picked up and transported by the autonomous passenger cars. The permission only applies to passenger transport. The operation of autonomous trucks is therefore not possible.

Waymo had already launched a robotic taxi service in the neighboring state of Arizona at the beginning of December 2018, the world’s first of its kind.

Virtual traffic simulation helps prep road network for automated vehicles

Virtuelle Verkehrssimulation (c) Lily Banse - Unsplash

Operation of automated vehicles poses new challenges to road network design. Vehicles are designed for ideal road conditions. Yet these are not available throughout Austria. A virtual traffic simulation should help identify and evaluate the most problematic routes.

Automated vehicles will drastically change the future of mobility. The Austrian government expects, among other things, heightened road safety, improved traffic efficiency and a reduction in energy consumption and emissions. Nevertheless, combined usage can be expected at first. Autonomous vehicles and traditional road users will share road networks.

Automated vehicle requirements

Before automated vehicles become operational, road infrastructure needs to be adapted. Smooth operation of vehicles depends on guidance systems, clear ground markers, clearly legible traffic signs and unrestricted visual connections. Austrian road infrastructure covers one hundred and ninety thousand kilometers; a comprehensive adaptation would be costly.

A research team from the Austrian Institute of Technology (AIT) in Vienna have explored the question of what infrastructure automated vehicles need in order to be able to travel safely and efficiently on Austria’s roads. The project, called Via Autonom, focused on non-urban roads, i.e. motorways, highways and open country roads.

Identification of vulnerabilities

The project aimed to identify weak points in road infrastructure. Central to the research were issues on safe and efficient combined usage and improvement of autonomous driving functions. Autonomous driving capabilities are actions such as lane planning, anticipatory driving style, pre-crash support and expanding the electronic horizon.

Finally, it was necessary to identify road sections where measures had to be taken so as to avoid risks in terms of road safety, traffic flow and driving comfort.

Risk modelling method

A method of risk modelling was developed in order to identify critical sites and sections of the road network. The focus was on linking sections, construction sites, unclear curves, etcetera. Building on this, analyses were carried out using traffic simulation methods. What has been analyzed:

  • the effectiveness of a predefined portfolio of measures;
  • the availability and quality of various data sources in terms of road safety and traffic flow;

“We used two demanding standard traffic scenarios in the virtual traffic simulation to examine what state-of-the-art vehicle sensors can do at the moment,” explains Philippe Nitsche, AIT project manager at AUTONOM.

Virtual traffic simulation

These are the tasks that the automated vehicle had to perform:

  • turn left at a secondary T intersection;
  • drive over a two hundred meter long acceleration runway into highway traffic.

The virtual traffic simulation was based on the assumption that the vehicle would be driven in accordance with road traffic regulations. Nevertheless, the automated vehicle could not satisfactorily handle both test situations. The results were sobering:

  • Only fifty-two percent of the eight hundred simulations on the short acceleration lane were successful;
  • When turning left at the T intersection, just forty percent of six hundred simulations were successful.

Improving infrastructure measures

Infrastructure measures which were implemented during the traffic simulation significantly improved the performance of automated vehicles. For example, the acceleration lane was extended and a cross-over assistant was installed at the T intersection.

Traffic simulation as a basis for making decisions

The study demonstrated the quality of virtual traffic simulation. The procedure is suitable for quantitative evaluation of infrastructure measures and their impact on safety and traffic flow. The tests are reliable and can be used as a basis for investment decisions.

Dynamic route planning

Navigation of automated vehicles is dependent on dynamic data. Dynamic data means real-time information about situations such as snowy roads or construction sites. If autonomous driving is no longer possible, controls are handed back to the driver or an emergency stop is initiated. The limitations are defined in the Operational Design Domain of the car. A prerequisite, however, is the permanent availability of information about current road conditions. Provision of this data requires cooperation between public authorities and commercial data providers.

The results of the project were:

  • a set of recommendations on infrastructure measures for autonomous road transport;
  • an identification method for critical sections within the Austrian road network;
  • an architectural concept for the efficient use of traffic, vehicle and infrastructure data along with digital maps;

Project partner:

The project was carried out in cooperation with business enterprises. The partners were IT service provider Prisma Solutions, traffic planners Rosinak & Partner, Virtual Vehicle R&D Centre and Wieser Verkehrssicherheit GmbH.

Also of interest:

Autonomous Driving: Solutions for road traffic of the future

On your own in the asphalt jungle

Pegasus – Mit dem Wunderpferd zum autonomen Fahren

Blickfeld: Deutsches Start-up entwickelt serientaugliche Schlüsseltechnologie für autonomes Fahren

Autonomes Fahren: Mit PATHFINDER auf den richtigen

WegErster automatisierter Lkw-Konvoi auf öffentlicher Straße

What robot fish, bees and self-driving cars have to do with each other

“Don’t talk too loudly or move too fast”, whispers Professor Tim Landgraf from the Freie Universtät Berlin. “Otherwise the fish will freak out.” He has just pulled up a large curtain which is surrounding a big tank with four small fish. Well, there are three of them, because one of the fish is actually a robot that is moved back and forth with the help of a magnet underneath the aquarium.
The goal, Landgraf explains, is to make the robot fish as much as possible a part of the group of the real fish in order to map the behaviour of the school of fish. “We’re getting better and better at it,” he says. The robot fish is becoming more and more accepted as a group member. It was a matter of taking small and bigger steps. When the fish got realistic large round eyes instead of painted dots, it proved to be an enormous leap forward. “That’s what the other fish were scared to death of.”

Professor Tim Landgraf, photo FU Berlin

It is just one of the nature research projects that Landgraf is working on at the Berlin University. The other major project involves a bee population in which all bees have been tagged so that their behaviour can be observed over a lifetime.

But why? Why is that information of any use to us?

A bee with a traceable tag on its back, photo FU Berlin

Food exchange

Landgraf does a lot of thinking. Animals and plants are often extremely good at particular things. The way in which bees collect and search for food, how they communicate with each other, etcetera. These are processes that have been perfected over hundreds of thousands of years. If you can capture that in algorithms, it can be extremely educational. “Everything we see in nature is enormously complex and are actually highly developed technologies from which we as human beings can learn if we look closely enough.”
An example of this for bees is how they use their energy. Like humans, bees need food to carry out their work. But sometimes their energy reservoir is empty, while the work is not yet finished and there are no flowers with nectar nearby. Bees have found a solution to this problem. They have a kind of second stomach with reserve food for friends in need. If one of the bees runs out of food, it can fill up on food from another bee.

Cars that refuel each other, photo FU Berlin

Auto-pilot driving

It’s a concept that Landgraf says people could use with electric cars. Especially when all cars will be driven using autopilot. You could provide them with a spare battery that would help other cars which have a flat battery. This would partly solve the problem of the limited range of e-cars. The refueling can even be done while driving, says Landgraf. Then you won’t waste any time.

And that just happens to be another research field of research at the FU Berlin. Twenty meters from Landgraf’s office, other FU employees use robot cars in their quest to perfect autopilot driving. Docking is also being tested. Outside, of course, there is also a real car to try out in the real world what has been tested in the laboratory set-up. Different fields of research intersect, says Landgraf. According to him, it’s the sort of research that Facebook and Google also do. They just have a little more money.

Tomorrow is Good: a choice guide for autonomous cars?

autopilot Tesla

A while ago, I attended a research presentation on autonomous cars by researchers at the Stuttgart University of Applied Sciences (DHBW). I learned that there are different levels of autonomous driving.

The most basic level is ‘hands-on‘, in which the human being controls the car completely by himself and is supported by the car while driving, for example by adaptive cruise control. A higher level we have ‘hands-off‘. The car can steer itself, but the driver must be able to take over the steering wheel at any time. It is important that the driver keeps his eyes on the road. In practice, this turns out to be difficult. The researchers at the Stuttgart University of Applied Sciences discovered with the aid of mobile eye tracking that most drivers already let their eyes wander after a few seconds, despite the fact that they were explicitly given the task of keeping their eyes on the road. Level 2 is, therefore, a level that we better skip.

On to level 3: ‘eyes-off‘. At this level, the car is programmed in such a way that it is no longer necessary for the driver to constantly focus on the road. Basically, the driver can sit back and read a book. Should a critical situation arise, the car will inform the driver, who can immediately take over the wheel.

autonomous driving Levels of autonomous driving (developed by the Society of Automotive Engineers)

Only in the upper two levels, we see the full range of autonomy. No driver intervention is required; the cars are programmed in such a way that they can do everything themselves. On level 4, ‘mind-off‘, there is still a steering wheel, but on level 5, the buyer of the car can also simply choose to leave out the entire steering wheel. In fact, on level 5 the whole driver has become superfluous. The cars on levels 4 and 5 have been morally programmed in such a way that they can make their own choice for every situation, including those on the interface between life and death.

Of course, this is the (near) future. The question is not so much when, but rather how. According to the Stuttgart researchers, Tesla’s autonomous cars are programmed in a more risk-oriented way than Daimler’s cars. The Daimler’s risk-averse moral programming ensures greater safety, not only for yourself as a “driver”, but also for other road users. Nevertheless, Tesla is more popular with most test drivers at Stuttgart University of Applied Sciences, not because of its moral programming, but rather because of its attractive user interface design.

Design, therefore, seems to be winning over moral programming. That made me think: will the way in which technology is morally programmed ever be part of consumers’ buying arguments? Will the moral programming of smart technology become part of the elevator pitch of an average company? In a few years’ time, will we hear consumers say ‘I have chosen the autonomous Mercedes because the way it is morally programmed fits better with my principles’? Or is there perhaps going to be a choice guide that will help us to interpret the differences in moral programming, just as it now helps us to interpret the differences between political parties? I’m really curious!

PS for everyone who is interested in all modern mobility services, including autonomous driving: come to the automotive campus in Helmond on Sunday afternoon, June 2 to the Mobifest: all interesting new mobility developments on a public day, prior to the ITS Europe Congress that will be organized next week in Eindhoven and Helmond.

Mobifest Helmond

About this column:

In a weekly column, alternately written by Eveline van Zeeland, Jan Wouters, Katleen Gabriels, Maarten Steinbuch, Mary Fiers, Carlo van de Weijer, Lucien Engelen, Tessie Hartjes and Auke Hoekstra, Innovation Origins tries to find out what the future will look like. These columnists, occasionally supplemented with guest bloggers, are all working in their own way on solutions for the problems of our time. So tomorrow will be good. Here are all the previous episodes.