“The groundwork has now been laid for a big step forward,” says Harry Wagner, Professor for Automotive and Mobility Management at Technische Hochschule Ingolstadt. He is referring to the legislation enabling Level 4 automated driving in Germany. A law passed last year already permits this in certain controlled areas, such as on a campus or to provide transport at an airport. These Operational Design Domains (ODDs), as the industry calls them, are chosen based on factors such as traffic, road features or weather conditions. That legislation was supplemented in May this year by an ordinance designed to amend road traffic law so as to create a consistent legal framework for autonomous driving. It is intended to keep up the pace of development and act as a stopgap until an agreement is reached at European level.
Autonomous Driving: Taking responsibility
Germany has established the legal framework for Level 4 fully automated driving. There is still a great deal to do on the technical side, though – from functional safety to software and artificial intelligence and on to camera technology.
By Markus Strehlitz
More robust hardware and software needed
Germany is working hard in this area, and lawmakers want autonomous vehicles to be an option for public transport, too. But on highways and in cities, Level 4 autonomous vehicles are still very much at the test stage. Like other experts, Wagner knows that the legal side is just the basis for progress rather than progress itself. This is difficult to overstate. In reality, the move to Level 4 is more of a leap than a step. It is a much bigger deal than the steps from Levels 1 to 3 (see box), since it is only at Level 4 that a vehicle takes over every driving task independently. “A Level 4 car still has a steering wheel, gas pedal and brake pedal,” says Wagner, but it does not need anyone to operate them. Only when the vehicle reaches its system limits, for example when it leaves its ODD, does it hand control back to the human driver.
That also has implications for questions of liability. At Level 3 and below, the driver has full responsibility in the event of an accident. But at Level 4 that is no longer necessarily the case, says Wagner. “Then you as the driver could say to the manufacturer, ‘Your system didn’t work; you’re liable for the damage.’ And, in some cases, we could be talking about a lot of money.” That makes functional safety vastly more important from Level 4 upwards. There needs to be considerable improvement in the robustness of hardware and software. And the amount of software in the vehicle is considerably higher again than at Level 3.
Germany’s new legislation on automated driving also encompasses functional safety. It requires manufacturers to present a general safety concept covering the vehicle’s functions and the information technology used. A sticking point here is the increased use of artificial intelligence. This is needed so that the car can perform the tasks intended at this level of automation. It enables the vehicle to detect and understand its surroundings, for example, and then to make decisions on what to do next.
It is the first of these – detecting surroundings – that is a particular challenge, says Josef Jiru, who is responsible for business development at the Fraunhofer Institute for Cognitive Systems (IKS). With traditional software, testing can provide a reliable safeguard, but this is not possible with AI systems. A system may interpret an image correctly every time during testing but then come up with a different result in the real world – if even a tiny bit of interference enters the frame, for instance.
Five steps to autonomous driving
Level 1: Driving assistance
Level 2: Partial automation
Level 3: High automation
Level 4: Full automation
Level 5: Autonomy
Wrangling over the best technology
Efforts are being made to tackle this challenge. Fraunhofer IKS, for example, is working on structured safety analysis. As Jiru explains, this creates a logical model of the system architecture that captures information flows and their quality along with limitations of the sensors. It is then possible to analyze how critical the identified weaknesses are and the risk they pose. “The safety analysis examines which critical situations lead to safety-relevant errors,” says Jiru.
He also describes another approach, which uses intelligent cross-validation of sensor data. This compares the data from one sensor – for example the front camera – with the information from other sensors that have different weaknesses, such as the lidar system. The sensors can then effectively check each other’s plausibility,” explains Jiru.
Harry Wagner also has cameras and sensors in mind when it comes to hardware robustness. Components such as cameras, lidar systems and actuators must be harmonized with each other, and getting there is no trivial task. “Beyond the software, we also need failsafe hardware,” says Wagner.
A further complication is that the car industry has not agreed on which approach is the right one for detecting surroundings. Manufacturers like Volkswagen use a combination of cameras and lidar sensors, while Tesla relies on camera technology alone.
Alliances are now being forged to drive development in this area. Bosch and VW, for example, aim to jointly develop a 360° video perception software to combine data from cameras, radars and sensors and process them with AI. Mercedes-Benz is working with US-based Luminar, a specialist in optical distance and speed measurement. The partnership aims to further develop lidar technology. This process is an ideal starting point for the use of binding standards, believes Dr. Ralf Petri, Head of Mobility at VDE. “We need to promote innovation – but for market access and penetration we also need standardization.” The fact that Level 4 driving is legally restricted to clearly defined ODDs should make development easier, he believes. He expects the technology to be used on highways first. “That’s the simplest scenario. There are no pedestrians, no bikes and no traffic lights.” Next might be rural roads. “The further you get into the city, the more complicated it becomes.”
Christian Hartmann agrees. He is Audi’s media spokesperson responsible for automated driving. “Level 4 driving in mixed traffic in the city, perhaps with a delivery vehicle blocking the way or a dog running into the road – that isn’t going to happen.”
Flexible driving means moving beyond set levels
Hartmann expects that Level 4 driving will always require vehicles to communicate with the local infrastructure in the ODD, just as a pilot communicates with air traffic control. In any case, Audi is looking to move away from the rigid division into five levels of automated driving. “Our approach is that there will be a mix of the different levels in the future so that the driver always gets the best support.” The vehicle will offer the driver different functions depending on the circumstances of the current journey. “For a tricky merge onto the highway, for example, a well-functioning Level 2 system, such as lane-keeping assist, could make sense.” Once that hazard is out of the way and the conditions are right for Level 4, more functions would then come online. “The vehicle will first check that all the conditions are satisfied. Only once that is the case will the Level 4 system become available,” says Hartmann.
In other words, the vehicle itself will recognize when it has reached a certain ODD and then offer the relevant functions. That is also the key difference from conventional assistance systems, Hartmann adds. When can people expect to see Level 4-capable Audi vehicles on the road? Hartmann asks for patience. “It won’t be until the second half of the decade under the current road map.”
It is generally difficult to judge how much progress individual manufacturers are making in developing their automated driving technology, says Wagner, noting that automotive companies are very reticent to communicate this information. That said, it is clear that manufacturers are working to overcome the various challenges and enable Level 4 driving.