Robotics & Automation: Driving towards an autonomous future
In this fast moving world people are easily distracted, especially it seems when driving. Behind the wheel for a few hours we might become sleepy or bored, perhaps we scan through the channels on the radio, attempt to drink a cup of coffee or eat a sandwich, fiddle with the car navigation system, or turn around to tell the kids to behave.
Worse still we drive after having a few drinks, we speed, we send text messages, we drive aggressively and break the rules of the road when no one is looking.
The World Health Organization estimates that 1.25 million people are killed on the road every year and more than 20 times that number are injured in traffic accidents. Estimates suggest that 94%1 of all road accidents (in the US) are caused by human error. It is likely that autonomous vehicles and cars with advanced driver assistance systems (“ADAS”) will provide a safer solution. The sensors on these vehicles will mimic our senses of vision and hearing, but they are also likely to surpass our human capabilities. Optical cameras, “LiDAR”2 and infrared sensors can be placed around the vehicle, eliminating the “blind-spot,” and enabling connectivity, to network the car with the infrastructure in the road such as stop signs and traffic lights. Communication with other vehicles will allow it to “know” that traffic ahead on the motorway or perhaps can even avoid dangerous situations when for example a car just round the bend on a country road has stopped before it enters line of sight.
“When automobiles first started rumbling down manure-clogged streets, people called them ‘horseless carriages’. This cycle has restarted, and the term ‘driverless car’ will soon seem as anachronistic as horseless carriage.”
Alex Davies, Wired Magazine, February 2018
Safety first, congestion second
Improving road safety to prevent human injury and the economic cost resulting from accidents is most often given as the first reason to push for autonomous vehicles. But there are other motivations. Technology companies see the opportunity to push further into a large potential market, and as the global population grows and cities become larger and more crowded, our infrastructure, the physical fabric of towns, cities and societies, is under pressure from over congestion. Autonomous vehicles could provide a solution.
A frequently cited estimate, although now 20 years old, suggests that cars are parked for 95% of their useful life,3 and this compounds with a different survey which shows that we use cars inefficiently, with approximately 80% of all motorway journeys being individuals driving alone.4
Autonomous vehicles, if not carefully regulated, could in fact aggravate congestion further. It might for example be cheaper to keep your autonomous car on the road with no human passenger at all, hopping between parking spaces to avoid traffic wardens, rather than paying for a parking space in your apartment or office building. However, a more positive outcome is also plausible. In this scenario, cars are shared or pooled and connected to an intelligent network to optimize traffic flow, route the supply of available vehicles towards demand and away from bottle necks. The OECD estimates that a system comprised entirely of autonomous networked vehicles operating in optimal conditions could meet today’s demand for passenger car transport (currently served by approximately 1.28 billion cars5 worldwide), with a fleet just 3% of the size.6 This outcome however, due to different interest groups, seems unlikely even in the smallest towns and cities. McKinsey estimate that the most likely scenario is that car sharing of autonomous vehicles will marginally offset the increase in vehicles in circulation (Figure 1). BP plc, clearly a company with a “pro-oil” bias, estimates that the number of cars on the road will grow to reach 2 billion by 2040.7
Automating the automobile
30 years ago the sunroof on my mother’s VW Golf operated by winding a handle, the windows likewise. We would open and lock the vehicle by inserting a metal key into door, adjust the seats with handles and levers, and keep a collection of maps in the glove box in case we lost our way. Many cars today are highly automated: automatic transmission, automatic headlight settings, rain sensors in the windscreen wipers, seat-memory which recognises the individual key fob, infotainment systems which connect to our smartphones, and of course the all important GPS navigation system.
Today we have become so accustomed to the automated conveniences in our own cars that renting a more basic vehicle on holiday can present some surprising challenges. How difficult it now seems to park, especially in an unfamiliar rental, without a rear-view camera or “park-assist” alerts. How embarrassing to stall the engine at every junction forgetting to change into neutral or depress the clutch. And where is “reverse” on this gear box?
Technology has been advancing along an exponential path and is the driving force behind the digitalization of our daily life and our increasing use and reliance on automation and robotics. In 1965 Gordon Moore, the co-founder of Fairchild Semiconductor and Intel, predicted that the number of transistors on an integrated circuit (or semiconductor or “chip”) would double every year. In 1975 he tempered this forecast to doubling every two years, but this prediction has held remarkably true and it describes a stark acceleration in computing power and a significant reduction in cost.8
Today we are seeing these new technologies entering the car. Enabled by sensors, powerful computer processors, a new generation of “artificial intelligence” and connected by fast mobile networks. The amount of technology being designed into vehicles today is growing fast and this increase in digital content is changing the structure of an industry and bringing myriad new players, very large and very small, into an industry now more than 100 years old.
So many new innovations are now being developed for and added to new models that it is difficult to keep track of the number. Japanese sensor and robotics company Omron, has developed an ADAS system which monitors the driver’s eye movements and “blink rate” in order to warn of drowsiness (Omron’s “Driver Concentration Sensing Technology”). Australian company, Seeing Machines, has developed a technology which goes a step further in tracking the driver’s pupils to check that the driver is aware of all surrounding hazards. Mercedes and Tesla has active “lane keeping” and active “blind spot assist” for motorways, the BMW 5 and 7 Series have infra-red vision as an optional extra for enhanced night time and poor visibility driving, and autonomous emergency braking (“AEB”) will become mandatory on all new cars in the US in 2022, although remains an option in Europe. Many car makers and suppliers are pushing into these areas as a means to differentiate their models against the competition and to add to the ever growing list of lucrative optional extras available to the customer.
Degrees of autonomy
Most of the technologies mentioned are used to automate part of the driving process or to augment a person’s skill in driving a vehicle. The Society of Automotive Engineers (the SAE) has defined a spectrum ranging from basic driver assistance systems, such as parking sensors, as “Level 1” automation, all the way up to Level 5 “full automation”, where a vehicle can perform all driving functions in any conditions, and driver interference is entirely optional.
Many billions of dollars have been invested in the development of the technologies and assistance devices used in autonomous vehicles today and many more will be invested in the next 5 to 10 years. Today the Audi A8 is the only vehicle on the road with “Level 3” autonomy. The A8’s “traffic jam pilot” can control the vehicle on a motorway with complete autonomy up to the speed of 60kmph. Beyond 60, the driver must take control (Figure 3.).
The challenge in achieving level 5 autonomy is enormous. In a controlled environment, such as Nissan’s geo-fenced area in Minato-Mirai, Yokohama, where they operate the “Easy Ride” service, or in Singapore’s One-North business district where MIT born nuTonomy, now a Delphi subsidiary, operates a fleet of robo-taxis, the number of variables in road condition and street signs, and possible scenarios, can be controlled to some extent. However, when in “the wild”, outside that controlled environment, that number of variables escalates fast. Developing a system which can immediately recognize all objects, differentiate between critical hazards (eg. people, animals, etc.) and much less significant noise (litter, puddles, fog, etc.) and respond appropriately in every scenario, and do so with a system with limited battery, and confined in physical dimensions and weight, is likely to take many more years of development.
The race has started. Companies are building out their fleets in order to expose them to as a many of these real world scenarios as possible. Networked, these autonomous systems have an advantage over human learner driver counterparts. The data collected by each of the autonomous vehicles can be pooled and the knowledge shared and learnt across the fleet. Tesla claims to have logged more than 1 billion miles of data with their vehicles operating in “Autopilot” mode. Tesla’s Autopilot is, however, only level 2 autonomy. Google subsidiary, Waymo (set up as a separate company in December 2016), Uber, the private company disrupting established taxi companies the world over, and Cruise Automation, owned by GM since March 2016 are the leaders to date in autonomous mileage. Waymo claims to have collated 4 million, Uber 2 million and Cruise 0.5 million miles at level 4 autonomy.
This picture could change fast however. Companies are now racing to expand their autonomous fleets in order to amass deepest and richest datasets. Earlier this year BMW announced they will double their self driving fleet to 80 vehicles in 2018, and Apple had registered 62 vehicles in California in May, up from 27 in January. But plans by Waymo and Uber have more recently raised the ante: Uber announced in November 2017 an order for 24,000 Volvo XC90s for delivery between 2019 and 2021. And in May 2018, Waymo placed an order with Fiat Chrysler for 62,000 mini-vans to their robot taxi fleet.
To err’ is human, but sometimes also robot
In March a car operating with Uber’s autonomous driving system caused a fatality in Tempe, Arizona. Autonomous vehicles have been involved in a number of accidents, but this was the first time a person has been killed. Most of the accidents to date have involved cars driven by people who did not anticipate the level of caution and hesitation displayed by the autonomous system.
This raises a dilemma. How many fatalities is society and the regulator willing to accept in the development of autonomous vehicles. Most experiments, especially ones which may be dangerous, are conducted in controlled environments and with people who have actively volunteered and are often paid for putting themselves at risk. From the social and media reaction following the accident, it is clear that people are far more forgiving of human error, “we all make mistakes”, than of a system error in a computer.
Other critical issues also need to be addressed. A car with autonomy levels 2-4 requires that the driver remains vigilant and capable of taking control if necessary. However, the need to remain vigilant and ready to take control, but not actually driving is likely to be more difficult and possibly and sleep-inducing than actually driving. Later down the road, if full autonomy is sold as part of the car’s functionality then would accidents qualify as product failures and will the car company become liable for damages? Audi currently assumes all liability for accidents caused by the A8 model.
In the next few years we expect that governments will step in with infrastructure investments to simplify the path to fully autonomous vehicles. Their incentive will be to reduce congestion and pollution and improve road safety. Many of the challenges of developing effective and safe autonomous vehicles could be reduced by making the surrounding infrastructure “smart”.
Today’s autonomous vehicles are designed to “read” and interpret street signs and road markings optically, but if the fabric of the road infrastructure were digitalized by embedding chips and sensors, then smart traffic lights and street markings could interact with the vehicles and make the technical challenge significantly simpler. Perhaps people might also place a digital tokens in their wallets or they might be embedded into smartphones, pets could have a token attached to their collar, and bicycles and other assets such as non-autonomous vehicles could all be tagged, digitally connecting them to the network and making the entire system safer and more efficient.
Street infrastructure and vehicles might also start to incorporate features to make them more easily identifiable to autonomous vehicles. PPG Industries, a leader in automotive paint, has developed coatings to improve the visibility of vehicles and infrastructure to radar and LiDAR sensors. The paint allow lights to penetrate to a reflective under-layer. The signal bounces off this layer and returns to the sensor instead of being absorbed.
In such a networked traffic system the possibilities for automation technology become myriad, but at the same time the requirements for improved cybersecurity escalate. Networking so many assets will create billions of end-points which need protecting from hackers, malware and viruses. We believe that cybersecurity is another compelling thematic opportunity for long term technology investors.
Only in the last few years the concept of self-driving cars has moved in our collective perception from being a possibility, to becoming a near certainty, with the only real question being when will they be ready for mainstream use? Many of the pieces of the puzzle, from regulation, through technology and social acceptance, still need to be carefully worked out, and before we reach the mass adoption of autonomous vehicles, we will no doubt witness ever more advanced ADAS systems, and suffer many more accidents and more public skepticism.
As technology advances and processing power increases, automation systems and robotics will become smarter, self-healing and ultimately more useful in an ever wider range of use cases. We are already witnessing this process in many areas of the economy, such as factories, hospitals, offices and even our homes. The car provides another good example of the steady rise in automation. When automobiles were first introduced they were known as “horseless carriages.” Perhaps the name “autonomous vehicle” or “self-driving car” will also pass, and over time these fully autonomous vehicles will simply become known as “cars” again.
Credit Suisse Asset Management have designed a strategy to provide clients with “pure-play” exposure to the compelling and multifaceted long-term secular growth theme of robotics & automation, which includes exposure to autonomous vehicles. For further information please click here.