By: Hans Adlkofer, Vice President, Automotive Systems, Infineon Technologies AG
There are several steps from today’s vehicles, with advanced driver assistant systems (ADAS) to fully automated self-driving cars. The automotive industry has already reached the first levels with partial automation. But there is still a long way to go towards highly automated cars and finally fully automated driving cars, which is not expected before 2025. While functions like adaptive cruise control, autonomous braking or parking assistance are already deployed, the road to fully self-driving cars requires a lot of further learnings and innovations but also correct legal framework. The article describes from a semiconductor perspective the different levels that are necessary to make completely self-driving cars a reality and also discusses related system requirements and the security aspects of connected cars.
According to the Society of Automotive Engineers (SAE International), road vehicle automation can be classified into six different levels. While the German Association of Automotive Industry (VDA) defined a similar classification, there are however, other definitions for vehicle automation levels available and commonly used in practice. The National Highway Traffic Safety Administration (NHTSA) in the US for example uses five different subclasses instead of the six.
The development from cars that are fully operated by the human driver to completely automated driving systems will be realized step-by-step with different automation levels (figure 1). According to SAE, Level 0 is completely under control of the human driver. Level 1 is characterized using driver assistance. Level 1 is already accomplished with driving mode-specific execution by a driver assistance system. Adaptive cruise control (ACC) and autonomous braking are typical examples of Level 1. These kinds of systems either execute the steering or acceleration/declaration using information about the driving environment but the human driver still performs all remaining aspects of the dynamic driving tasks. Level 2 (Partial Automation) sees driving mode-specific execution of both, steering and acceleration/declaration. With park assistance systems, ACC and lane keeping control Level 2 is already implemented in modern cars. With Levels 0 to 2 the human driver still monitors the driving environment and can have full control of the car, if desired.
Above Level 2, the automated driving system monitors the driving environment; it interferes with the driver’s actions or even takes over the control of the car for specific tasks or actions. Reaching Level 3 (Conditional Automation) an automated driving system controls all aspects of the dynamic driving tasks with the expectation that the human driver will respond appropriately to a request to intervene. On this highly automated level there is still a fallback to the human driver. Nevertheless, with applications such as traffic jam control, highway driving, automated parking and road train it is a big and challenging step to reach Level 3.
With Level 4 (High Automation) we will come to a highly automated automotive system, where there is no need of a human driver response immediately. And then on Level 5 (Full Automation) the fully automated driving system (self-driving car) takes care of all aspects of the dynamic driving tasks under all road and environmental conditions typically managed by a human driver today.
According to the latest report “Autonomous Driving: Question is When, Not If” (an update to a report of early 2014) of the market research company IHS Automotive, various automakers have increased their activity and investments toward the goal of self-driving vehicles.
OEMs remain geared toward augmenting the driver and adding incremental autonomous functions as autonomous driving technology improves. These findings further support the IHS Automotive global forecast for nearly 12 million in annual sales of self-driving cars in 2035. They also outline longer-term opportunities as nearly all vehicles on the streets today are likely to become self-driving cars or self-driving commercial vehicles on some level sometime after 2050.
To help with development, many testing areas for self-driving cars were established in 2014. Vehicle manufacturers are working with industry organizations, suppliers and university research conglomerates on these efforts.
IHS Automotive anticipates low-speed Level 5 self-driving vehicles could enter volume deployment in 2025 with full deployment of L5 self-driving vehicles at any speed five years later.
To reach the higher levels on the way to self-driving cars a lot of developments are required. In general, the cars need a new electronic architecture with domain structure compared to the existing ECU approach. This new architecture has to be supported by high-speed internal data bus systems. To ensure a safe operation high redundancy in technologies and systems has to be implemented. In addition, self-driving cars need a variety of existing and new sensors. To handle the increasing data volume an increasing computing performance is needed. Additionally, we need up-to-date information about road condition, traffic situation etc. via a secure Car-to-Car communication with secure gateways. This external connectivity also allows in-field upgradeability of car functions. On top there is a need for innovative technologies to observe the human driver. And there are still a lot of many more learnings.
But technology developments alone are not sufficient. Regulation detangling and abandoning existing regulations is another challenge. One of the underlying regulatory hurdles is the Vienna Convention of 1968. According to Article 8 ”Every moving vehicle or combination of vehicles shall have a driver” and Article 13 ”Every driver of a vehicle shall in all circumstances have his vehicle under control…” there is practically no legal foundation that allows the implementation of highly automated driving in transport of humans and goods. Only assisted or partially automated driving would comply with this convention. This is implemented in road laws of almost all EU countries.
Rapid development of automation technologies caused a mention in Article 8 of the Vienna Convention in March 2014. According to the new amendment, on the one hand the driver still has to be present and also has to be able to take over the steering wheel at any time. On the other hand, the amendment allows the car to drive itself as long as the system “can be overridden or switched off by the driver”. Even though this represents a major step towards real application of automated vehicles there are still legal hurdles that need to be adjusted in order to apply vehicle automation on highways.
To create an automated driving system, there are different building blocks necessary. The vehicle dynamics and control need innovative solutions for braking, steering, acceleration, suspension and transmission. In addition, self-driving cars request redundant sensors and sensor technologies and multiple times more computing power than today’s cars, to get all information managed in real-time. Related vehicle sensors include radar, ultrasonic, cameras, laser, GPS and map systems. The sensor data have to be collected, merged and transmitted for central computing. Furthermore, external connectivity is key to enable a secure communication with the environment. This can be realized via DSRC (Dedicated Short Range Communication), Wi-Fi or cellular channels. On top of all these functions both powerful data processing and decision-making features have to be installed.
To reach the higher levels of self-driving systems two major steps have to be done. Typical driver assistant systems today collect data from a source (camera, radar, etc.) and process the data for a dedicated function using a related ECU with a dedicated algorithm. The results are displayed or used to control specific actuator. To realize a self-driving system, the car has to maintain a ‘picture’ of the environment around the car all time, a ‘picture’ about what the driver is doing and a status model of the car. This needs the fusion of the sensor data and redundancy of the control unit and algorithm
In a step 2, around the self-driving vehicle a “safety cocoon” (figure 2) has to be developed based on radar, camera, ultrasonic and laser sensors. It should implement functions like LDW (Lane Departure Warning), LKW (Lane Keep Assistant), FCW (Forward Collision Warning), BSD (Blind Spot Detection), HBA (High Beam Assist), TSR (Traffic Sign Recognition), BUA (Back-up Aid), etc.
In order to get from today’s object-based fusion systems to grid-based fusions (networks), some challenging system requirements have to be managed: High-end multicore CPUs have to be replaced by Matrix-GPUs with safety-guards MCUs. While today a memory capacity of a few MBytes is sufficient, in future about 200 MB to 500 MB will be needed for the grid history alone, while the software algorithms (matrix-based and floating-point) need additional 80 MB to 160 MB. Furthermore, the computing performance will increase up to 8,000 DMIPS and higher.
Safety and Security
To get to self-driving cars, we have to come from fail-safe to fail-operation systems. This means a high level of redundancy for the system design. All safety-related functions, including the power supply and the communication network, have to be redundant. Other safety aspects – to name just a few – are high diversification, fast fault tolerance, self-monitoring devices, multicore MCUs, watchdog sensing and finally ASIL D capable operating systems and functions.
The growing symbiosis between a vehicle and its environment offers plenty of opportunities for cross-vehicle improvements. These new technologies as Car-to-Car (C2C) and Car-to-Infrastructure (C2I) communication will strongly support the self-driving car. The benefits are obvious: Besides the increased driver comfort it is estimated that these technologies will help prevent up to 90% of traffic fatalities. Additionally, environmental information helps improve driving strategy and thereby reduce fuel consumption. Permanent wireless access to cars opens up possibilities for new business models, e.g. remote software updates that minimize costly recalls. But the accessibility of a vehicle from outside also significantly increases the risk (figure 3) of hacker attacks (e.g. via mobile phone, Bluetooth or Wi-Fi).
C2C and C2I communication are expected to improve road safety and traffic efficiency in the future. For example, drivers will be warned in case of road damage or car accidents ahead. The exchange of sensitive information, such as position and speed, is required. The integrity of this data needs to be protected because the car of the future will use it to trigger warnings and autonomous reactions. Additionally, protecting the driver’s privacy is important. This calls for a high level of security in many places and interfaces on the car.
Security solutions for onboard communication need to be safe and compliant with hard, real-time constraints and legacy bus standards, with minimum data overhead and costs.
Onboard security and secure communication address two threats. One is the manipulation of hardware and software for e.g. tuning purposes or in other kinds of fraud where a harmful hardware component or software does not fulfill the guarantied and defined function. The other threat is terrorism or cyber war, where the attacker has successfully conquered an Electric Control Unit (ECU), allowing to send or manipulate messages on the bus and interfere or disturb the defined functions to harm people or to simply steal the car.
Innovative semiconductor solutions enable the installation of automotive security systems. These kinds of security systems protect the user and car OEMs in the global market, assets, personal data and, last but not least, the lives and limbs of the users on the road.
Autonomous and cooperative driving leads to new vehicle concepts and new tasks for the driver. The automotive industry is on its way to fully automated self-driving cars. While the first levels are already reached there are still a lot of challenges to fulfill all demanding requirements. Innovative semiconductor and software solutions are key to enable self-driving cars to benefit from a safer and secure mobility.