With numerous players developing a wide range of automated driving systems, how can the industry get everyone on the same page? By Megan Lampinen
Automated and autonomous driving developments are gaining pace, with an increasing number of pilots making their way to public roads. But as innovation picks up, a potential problem is becoming clear: not everyone is using the same standards, or even the same language, in their testing and development practices. Testing protocols are currently very specific to an individual automaker’s systems and operational design domains (ODDs). In light of that variance, how can the industry make sure that everyone is on the same page?
Best practices
The US-based Automated Vehicle Safety Consortium (AVSC) believes that for safe and sustainable deployment of autonomous vehicles (AVs), developers must communicate information about the safety of their vehicles in a consistent way. “It’s a gnarly problem,” cautions AVSC Director Amy Chu.
The consortium goes about solving that problem through best practice guidance. The most recent best practice release, its eighth, lays out a common approach to the Evaluation of Behavioural Competencies. Designed to help developers evaluate the safety performance of their automated vehicles, it is based on an elemental set of behavioural competencies such as maintaining proper lane position or responding appropriately in work zones. “Linking behavioural competencies to a key set of scenarios for a given ODD provides relevant evidence for ADAS safety performance and increases confidence in the safety performance of an AV,” explains Chu.
One of the problems is that we do not have a common language to describe many of these system
A big part of best practice is to bring clarity on the terminology used to describe these behavioural competencies, and to the relationship between certain terms that are commonly used. Many common words may be used differently depending on which company is talking. Take ‘behaviour’ and ‘manoeuvre’ as examples. The AVSC describes ‘behaviour’ as a goal-oriented task that incorporates the various aspects of the dynamic driving task. It’s about the detection and the response. The ‘manoeuvre’ is part of that but more about the vehicle control, applying lateral control, acceleration or braking. “I’ve heard those two terms used interchangeably,” notes Chu. “Even within our group, we found we used them differently when we first started the conversation.”
Then there is the difference between ‘behavioural competency’ and a ‘behaviour’. The AVSC clarifies that the behaviour is what the automated driving system is doing, but the behavioural competency is the demonstrated acceptable performance of it. That would have pass/fail criteria and give developers a way to determine competency. “There are so many activities going on, and so many groups doing research and testing,” notes Chu. “We hope that this lexicon is adopted and used more commonly so that we can all speak the same language.”
Wider consistency
Other industry players have also been calling for standardisation of technology terms and definitions, and not just around AVs but also advanced driver assistance systems (ADAS). Some of these safety features have been available for 30 years but appear on the market under a wide variety of names. Research from AAA in 2019 found that there were 40 unique names on the market for systems that were all essentially Automatic Emergency Braking. There were 20 names in use for Adaptive Cruise Control, 20 for Surround View Cameras and 19 different names for Lane Keeping Assistance. In response, the AAA put forward terminology intended to be simple, specific and based on system functionality.
OEMs must be careful in the wording of their semi-autonomous or highly automated systems
Mark Rosekind, Chief Safety Innovation Officer at AV start-up Zoox and former Administrator at the National Highway Traffic Safety Administration (NHTSA), is similarly pushing for clarity on functionality in automated and autonomous driving. “One of the problems is that we do not have a common language to describe many of these systems,” he tells Automotive World. “If you are not clear about what is in the vehicle then it will confuse people. Tech companies have to be specific about what they are developing.
He sees technology today as falling into two clear buckets: SAE Levels 1-3 that require human responsibility for driving and Levels 4-5 that take on full responsibility and allow the human to sit back and relax. Rosekind suggest clarifying these two groups would be a good place to start to support the AAA’s efforts, and similar work by other organisations, to define a common language. “A common language is the first step,” he asserts. “The second is the establishment of some very clear education channels. You shouldn’t be able to drive a vehicle off a lot or have one delivered until somebody’s gone through all the systems.”
Adoption and regulation
A common language and the publication of best practices could smooth the development process among so many various companies, but other things are needed as well. “You need people to adopt these best practices and then present the data to show what’s working and what’s not,” says Rosekind. “It would be great to have more data on the first bucket, the ADAS part, around how big a difference they are making in terms of safety. Let’s demonstrate how good these systems are. If the full system is working, then federal regulators get into this and turn it into regulation.”
Linking behavioural competencies to a key set of scenarios for a given ODD provides relevant evidence for ADAS safety performance and increases confidence in the safety performance of an AV
The AVSC’s Chu notes that regulations can be a contentious subject. “The question frequently arises of which comes first: the regulation or the innovation? That’s always difficult to balance. I do not think that one can come first and then everything simply falls into place. It just will not work that way. Collaboration is key, and that’s one of the reasons we exist.”
The AVSC aims to accelerate the standards-making process through its best practices. “We end up taking the best practices that the group has agreed and walking them into standards committees and participating in various panels,” says Chu. Some SAE International standards have already been updated specifically based on AVSC documents. For instance, the SAE International made a special update to the J3018 (Guidelines for Safe On-Road Testing of SAE Level 3, 4, and 5 Prototype Automated Driving Systems) based on the AVSC’s In-vehicle Fallback Test Driver Procedures. “That was a win for us because what we’re aiming to do is accelerate standards-making,” she emphasises.
For Zoox, whose self-driving vehicle remains in development, much of its efforts right now are focussed on building up the public education piece. “There’s a disadvantage in the fact that we are not on the road yet,” notes Rosekind. “You cannot just take a ride and see how great the technology is.” Lacking that experience part, Rosekind and his team have committed to a series of safety reports around AV technology and aim to clear up some of the current confusion. “People are saying all sorts of different things. AVs can save lives but are we saving all those lives tomorrow? Every car will have this technology yesterday,” he notes. “We need clarity about all that and a consistent message. And not just from developers or manufacturers or the government; it has to be everybody talking about it the same way.”
Keyword: A common language and best practice key to AV development success