By Jonathan Agnew – Head of Insurance, recoveriescorp
There seems to be a mad rush by manufacturers and others, like Google, to deliver the first self-driving vehicle for consumers, but the recent fatal crash in a Tesla car on ‘Autopilot’ mode makes it clear that we still have a ways to go. While companies like Google and Ford are prepared to head straight for self-driving vehicles, the majority of car manufacturers have taken an approach of gradual progression, adding additional autonomous technologies to their vehicles over time; from cruise control and automated parking, to automatic braking at intersections and hands free lane change. Even Mercedes released a car that on its own - will slow, brake, and speed up again with traffic, but only in certain circumstances and for a very minimal amount of time. All of these features are available today to provide customers with an easier drive and more importantly, a much safer ride.
In 2013, the US Department of Transportation's National Highway Traffic Safety Administration (NHTSA) defined five different levels of autonomous driving, with Level 0 being non-automated, and Level 4 being fully automated1. The varied technology in the cars mentioned above falls between function-specific automation and limited self-driving automation (levels 1-3) by NHTSA’s classification2.
However, for many, Level 3, limited self-driving poses a significant challenge. Once a driver is allowed to relax and take their attention away from the car and the road, it can take a person up to 10 seconds to comprehend that there is an emergency, even with a cars internal warning system, such as flashing lights and verbal cues. Although details of the Tesla crash are still being investigated, this issue of transferring control from car to driver is so consequential that both Ford and Google have chosen to bypass Level 3 technology for consumers and focus on a fully automated vehicle that does not need human intervention to avoid an accident3.
Semi-Autonomous Technology to Self-Driving Vehicles
Today, plenty of the newer high end cars are fitted out with autonomous technology, such as GPS, radar, cameras and other sensors to give an accurate picture of location and surroundings. The equipment is used for both navigation and to detect and avoid obstacles. Teslas 2015 Model S, similar to the car involved in the crash, reportedly has one windshield camera and a radar sensor mounted low in the grille to view surroundings. Ultrasonic sensors check for a clear path to the side before enabling a lane change4. Unfortunately, and quite tragically, the cameras, radars and sensors, which all report information to the ‘Autopilot’ software, failed to recognize the white side of a tractor trailer against a bright lit sky, as reported on Teslas own blog. Essentially, ‘Autopilot’ didn’t know the truck was there, so it did not apply the brakes when the truck made a left turn right in front of it at an intersection5.
The accident itself has highlighted the limitations of current technology, but a 50 mile test drive conducted in four of the most advanced semi-autonomous vehicles on the market should also prove cautionary. The driver found technology lapses were triggered by damaged roads, unclear markings, sweeping bends, exit and entrance ramps, curbs, intersections, the general stop and go of city traffic and of course bad weather, all requiring the driver to take over4. While some cars performed better than others, there were still way too many disruptions for one to reasonably consider ever taking their eyes off the road and keeping their hands off the wheel, but people still do.
So where does this leave us?
Current law requires that all cars on the road must provide the ability for a human to take over, but Googles aim has always been to deliver a self-driving car, Level 4, where one is not expected to be available for control at any time during the trip. Google cars are also equipped with a camera system, radars and several sensors, but one technological difference, that just might be giving them the edge, is the Lidar.
Situated on top of the car, the Lidar spins around to give a 360 view, providing more than 1 million readings per second, of a distance up to 100 metres, compared to the average stereo camera which only views distances up to about 30 metres. This Lidar creates a detailed 3D map of the surrounding environment in real-time, while gathering additional information from other sensors and comparing all of this to existing maps so that it can identify any new obstacles or discrepancies. It has been described as very effective with a high amount of accuracy, but also very expensive at around $75,000 U.S. dollars for one6, until recently. A new Lidar has been unveiled, detecting objects at double the distance of the Google version and at reportedly only $500 per unit7.
I couldn’t tell you if a Lidar would have made a difference in the Tesla crash incident, however to be fair, even with the Lidar, and over 1 million miles of road testing, Google cars also have their faults and limitations. For example:
· A Lidar doesn’t accurately monitor the speed of the traffic around you, so Google still uses both front and rear bumper radars to assist with this.
· Difficult weather like heavy rain and snow are still an obstacle.
· On over 424,000 miles of road Google reported 341 incidents in which a driver needed to take over, of those 341 incidents Google acknowledged that 11 would have resulted in a crash had the driver not taken over the vehicle8.
· Earlier this year was the first reported incident of a collision caused by a Google car.
When Tesla decided to forego the Lidar, they began to hedge their bets that software and processing would be the game changer. Their camera system has software that can detect vehicles up to 230 feet away, to create a video stream that is analysed by algorithms trained to identify objects. This equipment allowed the company to track the movements of customers in Tesla cars and build a treasure trove of driving data, which would eventually be used to remotely activate ‘Autopilot’ software in tens of thousands Tesla vehicles that were already on the road9. Overnight, Teslas Model S owners had ‘Autopilot’ software which gave them some of the most advance autonomous technology available today. The catch? It’s in public beta testing, which means it’s not quite ready for the eyes off the road, hands off the steering wheel experience that Teslas early adopters are so eagerly waiting for and so many are too keen to try.
With this software Tesla collects more than a million miles of driving data every day, allowing them to continually improve the autonomous driving experience for their entire fleet of vehicles. The cars also contain a black box to help identify what happened in the situation of an accident. It is expected that with all this information, Tesla will learn from the recent tragedy and improve on their technology. Still, the death of Tesla driver, Joshua Brown, while in ‘Autopilot’ mode will definitely increase scrutiny in this area of technology, which is already heavily scrutinized - and could very possibly slow progress throughout the entire industry.
Moving Beyond Vehicle Automation
Fully autonomous vehicles are definitely on the way, but such a revolutionary move requires significant changes in laws, regulations and consumer/driver behaviour. In the meantime, as we integrate semi-autonomous vehicles on our traditional roads, with our traditional cars, it is expected that we’ll see more accidents like the one that involved Tesla. This is why it is the right time to look beyond just vehicle automation. A report by PwC suggest that vehicle manufacturers and traffic authorities are looking at two other types of technologies which will help aid society through what may be considered a more acceptable transition to getting fully automated vehicles on the road.
· Vehicle to Infrastructure Communications, such as automating road infrastructure like smart traffic signals to communicate with cars.
· Vehicle to Vehicle Communications, where cars communicate with each other10, making it possible to route traffic for efficiency.
It is these types of emerging technologies that could potentially solve some of the issues that remain unresolved with our current technology. For example, if smart traffic signals and/or tech savvy roads and laneways could communicate with our cars, it could potentially minimize or even eliminate the problems that come with bad weather. Some even suggest that the Tesla crash could have been avoided if both vehicles were able to identify and communicate with each other.
While some are eager to head towards fully autonomous vehicles and a self-driving vehicle society, there are others who caution on taking such a huge step forward too quickly. The truth is, the incremental progress in autonomous technology allows all of us the opportunity to get acclimated to an entirely new era of transportation.
In my next article we’ll look at underwriting in an era of driverless vehicles.
References
2. https://www.google.com/selfdrivingcar/faq/
3. https://www.wired.com/2015/11/ford-self-driving-car-plan-google/
6. http://googlesautonomousvehicle.weebly.com/technology-and-costs.html
7. http://www.azosensors.com/Article.aspx?ArticleID=688