Autonomous Driving

Autonomous or self-driving vehicles are being increasingly treated as a trailblazing technology of the future. The question is, what does »autonomous« really mean, and what will it take to make autonomous driving safe and efficient?

5 levels of autonomous driving

Most new automobiles are already automated in certain areas, beginning with standard-equipment driver assistance systems (DAS). In the future however, we will see completely autonomous vehicles on the streets.

Autonomous vehicles are usually separated into five levels based on the Society of Automotive Engineers classification system:

  • Level 1 driver assistance: This already includes automobiles with conventional cruise control systems.
  • Level 2 partial automation: The vehicle can change lanes or brake on its own in situations defined by the manufacturer. Parking assistants are also part of level 2 functions.
  • Level 3 conditional automation: Highly-automated vehicles can overtake, brake and accelerate on their own depending on the traffic situation. During this time, drivers can even read the newspaper, although they still have to be able to take control of the vehicle again if the system issues an alarm.
  • Level 4 high automation: At this level, the vehicle takes over all driving functions. This includes tasks such as entering a highway, activating turn signals and overtaking. At this level, drivers can even sleep while the vehicle operates autonomously.
  • Level 5 full automation: With autonomous vehicles, the passenger assumes no driving tasks and has no opportunity to intercede in the driving situation. The vehicle handles extremely complex driving situations such as autonomously traversing an intersection. These vehicles can also operate without passengers.
© iStock.com/Just_Super

Current status: How much autonomy is permitted on our roads at the moment?

To date, Germany has allowed only level 2 (partial automation) assistants. Considerable hurdles exist between level 2 and level 3. On one hand, faulty autonomous system behavior raises new legal questions. On the other hand, beginning with level 3 the system must be capable of monitoring the driving environment on its own and reacting to any changes. Extensive research is still required before this condition can be reliably satisfied. 

Opportunities and risks: autonomous vehicles have a wide range of applications

The fields of application for automated or autonomous vehicles are multifaceted. When the movement of people is involved, there are two conceivable scenarios:

  • shared autonomy through automated or autonomous taxis and buses
  • owned autonomy in the form of privately-used autonomous automobiles

Apart from the movement of people, autonomous driving will impact a wide range of other sectors of the economy, such as in agriculture, where autonomous vehicles and machines could reduce the demand for labor and increase efficiency.

Autonomous trucks could take over hazardous or monotonous tasks at freight ports or mines for example, or transport goods via highway convoys more efficiently from A to B without drivers.

Apart from the previously-mentioned legal questions that autonomous driving raises, a highly crucial issue is system dependability. When it comes to using autonomous systems in road traffic, human lives are at stake. Autonomous driving’s greatest potential can be exploited only if the vehicles operate error-free.

Technical requirements for autonomous driving

Level 3 and higher autonomous or automated vehicles must simultaneously meet a wide range of technical requirements. The three most critical are:

Machine perception

Cameras and sensors such as radar and lidar scan the vehicle environment in real-time. Digital maps and (Germany) GPS location systems offer additional information regarding the environment. Vehicles can communicate with one another using cooperative traffic systems and thus incorporate the planned driving behavior of the other vehicles into their own cognitive analysis. The collected data is merged and analyzed with artificial intelligence (AI), which results in the machine-based perception of the vehicle’s environment.

Situation comprehension

The next step involves the generation of a 3D model of the environment based on the vehicle’s perception. The system then calculates the imminent situation and generates forecasts. In the future, these individual steps, which are also carried out with AI technology, will run individually and independent of one another. This allows the system to analyze the result from time to time.

Lane maneuvering

The vehicle then plans the next activities and executes them autonomously. It remains on the ideal lane and activates other functions such as the turn signals or the brakes.

Dependable cognitive systems for autonomous driving

When it comes to autonomous driving, the greatest challenge involves generating and processing information, and then reacting accordingly.

Today, autonomous vehicles function reasonably well in test situations since the conditions are severely restricted and thus easy to manage. A key issue however is how to design autonomous vehicles so that they operate dependably even in difficult environments, such as normal road traffic.

If the system is not in a position to create a precise model of the driving situation in bad weather conditions for instance, the vehicle cannot be allowed to continue to operate. The system must be able to monitor itself and evaluate its own state and level of dependability while continuing to operate under these restrictions.

The Fraunhofer Institute for Cognitive Systems IKS offers solutions that permit autonomous vehicles to function dependably, in spite of difficult conditions or errors, thus ensuring that no one is exposed to danger. The goal is to create a thoroughly verified, intelligent software architecture for the automobile - a so-called resilient cognitive system.

© iStock.com/Tamas Gabor

Fraunhofer IKS validates artificial intelligence (AI)

Autonomous driving is based in large part on artificial intelligence (AI), machine learning and neural networks.  Because there is no possibility for human validation of the machine perception and the resulting decisions when using these technologies, other ways have to be found to analyze the accuracy of the machine perception. The Fraunhofer Institute for Cognitive Systems IKS conducts research into methods for validating the perception.

 One approach is the structured safety analysis, in which a logical model of the system architecture is created to represent the signal flows and their quality. The performance and limitations of the sensors are also described in the system architecture. The system then examines how critical these identified weak points are, including the associated risks, and then determines which critical situations lead to safety-relevant errors.

Another approach is the intelligent cross validation of existing internal and external sensor data. This involves comparing the data from a sensor with data from other types of sensors, each of which have different weak points. The data from the different sensors, such as a front-end camera and a lidar system, can then carry out a cross-check.

Safety through adaptive software architectures

© iStock.com/gustavofrazao

In order to validate autonomous driving systems, the Fraunhofer Institute for Cognitive Systems IKS also conducts research into adaptive software architectures. These architectures independently adapt to changing conditions in the environment, thus allowing them to proactively deal with interference factors.

The software-based functions in self-adapting software architectures are flexibly designed so that they can be shifted or operated without any restrictions, even when other parts of the system are experiencing problems. This type of software architecture is referred to as a fail-operational architecture.

In cases where the system fails to retain its full functional capabilities through adaptation, the function can be gradually abated through so-called graceful degradation, thus ensuring that the system remains safe and stable. In autonomous vehicles, this approach guarantees the flawless execution of safety-critical functions such as staying in the lane, even when components such as a camera fail.

Real-time connected traffic models for increased safety, efficiency and resource conservation

Another aspect of making autonomous driving safer is so-called Car2X communication, which requires equipping the vehicles and infrastructure with sensors. The goal is to create a cooperative ecosystem for road traffic, in which infrastructure and position data can be shared via edge and cloud computing. Cooperative driving and Car2X communication thus increase the efficiency, safety and sustainability of the traffic system. In this area, the activities of the Fraunhofer Institute for Cognitive Systems IKS are focused on the dependability of the systems, which can be easily and effectively increased with infrastructure data since the infrastructure sensors have a better overview of the critical traffic points than the individual traffic participants. Technically speaking however, that means the infrastructure has to be viewed as another undependable source of sensors in the E2E architecture.

Safe Intelligence Magazine

These are the latest articles on our blog on autonomous driving:

 

16.7.2020

Cloud-based systems face safety and efficiency challenges

The future will rely on widespread, massively-connected, highly-intelligent systems. To implement this vision, we need dependable cloud-based cyber-physical systems. Research in this direction encounters as many opportunities as challenges. Such systems exist in changing environments and interact with other systems and humans, which requires intelligence, autonomy, safety and adaptability – in fact, conflicting goals. An example of this is autonomous driving.

 

25.6.2020

Why Safety Matters

Machine learning means a disruptive challenge for safety assurance. It’s therefore no surprise that AI safety has gained much more attention over the past few months. This article focuses on the basic understanding of what safety actually means. It highlights why it is so important to understand that safe AI has less to do with AI itself and much more to do with safety engineering.

 

30.4.2020

Uncovering difficult situations for autonomous driving systems

Today, autonomous vehicles function reasonably well in test situations since the conditions are severely restricted and thus easy to manage. A key issue however is how to design autonomous vehicles so that they operate dependably even in complex and previously unknown situations. A solution from Fraunhofer IKS is helping to uncover and predict such difficult situations.