Exploring whether virtual testing is a realistic goal for carmakers by Morag Cuddeford-Jones. [Auto.Cuddeford.2016.06.22]
Autonomous and ADAS-enhanced cars are now a reality on our roads. Atelier.net predicts that most cars will have some kind of automated features, reducing the driver to a co-pilot status, by 2020. The Wall Street Journal reported that cars with autonomous features could have replaced the US’s current road fleet in 20 years.
With this growing presence we are already seeing not just the proof of concept but beginning to understand the scope of unintended consequences. Previously unpredicted accidents involving Google’s driverless cars as well as a Tesla fatality demonstrate that, as with all new technologies, extensive testing to mitigate risk is required.
Many carmakers are engaging in extensive road testing of their own versions of these new technologies. According to Autocar, Jaguar Land Rover announced in July 2016 that it would test its range of connected and self-driving vehicles on a 41-mile route across the Midlandsbefore the end of the year. But is there not a more effective yet at the same time less expensive way of testing the roadworthiness of these technologies?
For Tesla at least, it would seem that the world’s roads are already its living laboratory. According to company sources there are more than 50,000 of its vehicles on the road today. “If you have thousands of vehicles out there collecting data you have a good statistical understanding of what the normal behaviour of the vehicle is. This creates a deep learning process,” states Maxime Flament, head of department - connectivity and automation with Ertico.
Not all autonomous or ADAS-enabled carmakers have the luxury of an existing fleet relaying real time data back to HQ. In this situation, surely designing a virtual test bed will help inform design and accelerate launch while driving cost efficiencies? All would depend on the type of problem the automaker is trying to solve.
“If you are in a traditional, deterministic mindset where you want a vehicle to do exactly as it is told then you can do extensive tests,” Flament explains. “But when you’re talking about vehicles that should address all road scenarios, this mindset is very difficult to apply. Even if you test it against infinite amounts of data you would end up in situations where the vehicle still couldn’t cope.”
Nico Gollwitzer, head of telematics at Vodafone, believes that a virtual/real world hybrid will always exist: “A lot of crash simulation is done on computer but it’s based on the standardised crash test as defined by the EU commission. It’s a very controlled, clear environment. But even today, all the regulatory bodies still want to see a real crash. Where you can save money is being able to verify early in the design process what may or may not be practical.”
Francois Guichard , United Nations Economic Commission’s secretary for vehicle active safety - focal point ITS/automated driving, Europe, notes: “The only question is the quality of the simulation. If you can validate your model using simulation then clearly that simulation was valuable. You just end up creating science fiction otherwise.”
He adds that using simulation is vital when all cars are not created equal: “The cost of testing a stability function for a low production run vehicle is really high. Simulations make this much easier and are good enough to satisfy safety regulations.”
Gollwitzer states that the volume of data required from on road cars may put a strain on the vehicle’s connectivity. “From an infrastructure perspective that’s a huge challenge. Autonomous cars will need a constant data connection as well as local memory.”
There are clear stages within the design and manufacture process of autonomous and ADAS cars where virtual testing is of use. In rooting out design deficiencies and improving efficiency including wear and tear and ecology it can prove to be a very fast and cost-effective alternative to road testing.
However, carmakers need to be wary of using virtual testing to attempt to identify, define and resolve every possible scenario. It is a far more effective validation tool than a preventative one.“Where a party’s liability in the case of an accident is at question, product liability laws would automatically look to the standard of testing involved in design. Manufacturers are obliged to take all measures required that can reasonably be expected to be performed to ensure that their vehicles are safe. This allows the law to adapt to all sorts of circumstances,” reveals Dr Stephan Appt, partner at law firm Pinsent Masons.
In the case of virtual, digital testing of the capabilities and potential flaws of an autonomous vehicle, the result is that any testing equipment should be able to be considered ‘state of the art’ at the time the vehicle becomes available on the market.
“Technical experts really need to challenge themselves to find whether or not the increase in virtual testing is only a great way of cutting costs or if there is a benefit because you can test for more eventualities than you could in the real world. In the balance between the two, I would be wary of accepting it if the main driver [for virtual testing] were to cut cost,” Dr Appt concludes.
02 Nov 2016 - 03 Nov 2016, Munich, Germany
For 13 years, this event has grown enormously in size, scope and significance - totally reflecting the path that the connected car has taken from ‘concept’ to ‘reality’. To reflect how the future of the car is not only being defined by in-car connectivity, we have added two new areas of focus to our conference - new models of auto mobility and automated driving technology.