Drive thy self: A summary of the current state of vehicle autonomy.
Will self driving cars be common?
The romantic flirt and dance with the idea of self-driving cars has been a long standing and deep-seated human mobility desire. One of the ways in which we have seen this expressed is the art of film. This art provides a conceptual visual platform and peak into how society views the potential progression of vehicles driving themselves and even communicating with us such as in the 1982-86 action series Knight Rider. The perspective of film helps give a societal temperature check in terms of elements like technological expectations of the future all the way to societal fears (e.g. war, disease or terror) prevailing at the time. For example, we can see in Steven Spielberg’s 2002 movie Minority Report, a futurist view of what 2054 could look like and what that would entail for vehicles at this time. For this movie, Spielberg actually convened 15 experts ranging from computer science to architecture and bio-medicine in creating a view of what technologies 2054 would have.
With respect to mobility, film offers an interesting and unconstrained landscape to what autonomous vehicles could be like from design to size and other accompanying features. The key lesson in these depictions are that vehicles themselves are a technologically driven response to the lay-out of the real world environment and structure inhabited by humans e.g. high rise and densely populated urbanized cities.
Exiting the realm of film , we can take a look to what we see today to evaluate where we find the state of vehicle autonomy and how it is developing. Also worth noting is that the discussion around autonomous vehicles has been largely grouped with the discussion around electric vehicles which have been summarized in an earlier article here. This is probably because there is sometimes a perceived linearity in the manner in which they are developing. Though this maybe loosely the case, in this article, the key areas of focus will be highlighting the classifications of autonomous vehicles, the associated autonomy technologies as well as what it means for what the future could hold.
Levels of Vehicle Autonomy
The first layer of viewing the incremental dynamics of autonomous vehicle progress are the 6 levels of autonomy used to describe the extent to which a vehicle is indeed classified as autonomous. Some of the levels blur into each other without clean transition in practice since ultimately it depends on the exact features a car would have and how these features interact and combine. Therefore, we can rather first describe and exemplify what level of autonomy specific features are before concluding the combined autonomy level of a car.
Level 0 Autonomy: No Autonomy/ Full driver input
This is what we can define as the basic vehicle requiring a human to control all the aspects of driving the car i.e. steering , accelerating, braking, indicating, gear changing etc. From the times of horseback as the means for transport to the invention of the motor vehicle and it’s mass production by Henry Ford in 1908 in the form of the Model T, most road travel in cumulative history has been of a Level 0 autonomy nature.
Level 1 Autonomy: Driver Assistance
1958 Chrysler Imperial
At Level 1 autonomy, a vehicle still requires a driver for all the elements mentioned above in Level 0. The incremental difference in this case is that some autonomous features exist that aid the driver to a marginal extent. Most notable examples include park distance beeping and cruise control. Level 1 car features started getting popularized in the second half of the 20th century. The aforementioned example of cruise control is a feature which was first added to road cars in 1958 by Chrysler in its Imperial model. The actual name “cruise control” was then marketed by Cadillac soon after in 1960 and has since become the common term. Adaptive cruise control (ACC) is an added dimension to cruise control where the system uses vehicles in front for reference. On its own, ACC is still considered Level 1 autonomy. Mitsubishi was the first car maker to produce a road car with this adaptive cruise control in 1992 with its Debonair model and as the 90s progressed, ACC was seen in more cars across the spectrum e.g. the 1999 Mercedes S Class marketed by Mercedes as Distronic.
Level 2 Autonomy: Partial Autonomy
Partial autonomy enables drivers to disengage from some driving functions with a similar but heightened and combined capability to previously discussed Level 1 . At Level 2, the vehicle assists with functions like steering, acceleration, braking, and maintaining speed but the driver’s input is still mostly required therefore the driver’s hands are still needed on the wheel. An example of this is the combination of adaptive versions of cruise control with lane departure warning/ lane centering technology. Lane departure technology started getting popularized from the early 2000s in Japanese cars such as the 2001 Nissan Cima which was marketed as the Infiniti Q45 outside Japan. Today, lane centering technology is not an outlandish concept and is marketed by different manufacturers under different names e.g. Tesla’s Autopilot or Toyota’s Lane Tracing Assist.
2001 Nissan Cima
Tesla Autopilot at work
Level 3 Automation: Conditional Automation
At level 3 autonomy, the vehicle can drive without the driver’s input on the key elements of driving but only under certain conditions such as limited speed, freeway conditions or self-parking. When these conditions are satisfied, the car requires no driver input and theoretically the driver doesn’t need to be involved. However, a sudden change in these required conditions or an emergency , mean the driver has to be on hand to take charge. The 2019 Audi A8 was planned to be one of the first Level 3 capable cars however Audi has since abandoned adding this driver assist system to the 2019 model due to regulatory complications regarding government approvals and apportioning blame for legal liability in the event of a crash.
Revisiting the use of film, an idea of the Audi A8’s features can be seen in the video clip from the 2017 movie, Spiderman Homecoming which is part of the Marvel Cinematic Universe.
Level 4 Automation: High Automation
Level 4 entails much of what we see in Level 3 but to a higher degree with the vehicle able to do major functions from steering, accelerating, and braking on its own. At this level, the vehicle is also capable of monitoring and responding to obstacles in the process of managing these mentioned typical driving functions therefore allowing the car to turn and switch lanes on its own. This still requires a level of standard conditions with a manageable level of complexity. As things get more complicated and fluid e.g. dynamic traffic jams and major obstacles , level 4 automation would be challenged in successfully negotiating the situation though it could still be capable. The most prevalent example of work being done on Level 4 automation is the Google developed technology company, Waymo which car makers like Volvo and Fiat Chrysler intend to use in propelling their self-driving ambitions. Another example is the 2017 unveiled concept car by Renault called the Symbioz which has also attained Level 4 and whose technology Renault hopes to start incorporating into its cars from 2023 (2nd video below)
Waymo fully driverless vehicle demo
Renault Symbioz demo video
Level 5 Automation: Full Automation
Audi AI:CON Demo Video
Level 5 is the level at which completely no human input is required for all driver functions across all circumstances meaning that the person/people in the vehicle are all effectively passengers. Achieving this therefore requires the vehicle to compute vast amounts of real time data and therefore would make use of Artificial Intelligence(AI) to replicate and supersede human driving capability. Some estimations of data required to do this are put at anything from 4 – 15 TB/hour. In 2017 Audi displayed their AI:CON concept car at the Frankfurt Motor Show. The fully electric concept vehicle is Level 5 so there is no steering wheel or foot pedals and includes many other futuristic interface features.
The Main Technologies Involved
Having gone through the autonomy levels, the following segment is to summarize the key technologies involved in these developments.
Radio Detecting and Ranging aka RADAR, is a technology developed just before World War 2 which uses radio waves emitted by a transmitter in conjunction with the reflection of these radio waves from objects back to sensors. This reflection based measurement process then serves to detect the location of objects within the range. This supplements camera sensors that would be also involved in the vehicle mapping it’s surroundings since camera sensors require a decent level of visibility. Below is a schematic of the Level 3 Audi 8 which shows a slightly more detailed layout of the technology involved.
However, the technology currently touted to enable big steps in self-driving capability is Light Detection and Ranging aka LiDAR. This was developed in the 1960s as part of military and aeronautic applications and has since evolved to other uses such as seismology. Basically, the concept is similar to Radar but with LiDAR, light is projected instead of radio waves allowing a wider detection range. This light is as laser pulses from the car whose light reflection from objects is measured to render a 360 degree 3D rendering of surroundings.
Other manufacturers such as Tesla, believe that Level 5 autonomy can be attained without using LiDAR but through continuous software updates on top of the current computer vision and neural networks software that Tesla cars use. This view has proved quite controversial in terms of how possible that would be especially within the time-frames communicated by Tesla therefore it remains to be seen whether this can be indeed achieved.
If it is to be the case, the cost limitations of LiDAR would be greatly mitigated. Granted though, like most technology elements, the cost of LiDAR is coming down. As an example, Silicon Valley based company Velodyne which is one of the market leaders for LiDAR sells its most popular package at $4000 per unit. Waymo used to purchase their LiDAR package from Velodyne at approx. $75 000 but have since reduced costs significantly by bringing the development process in-house to the extent that as at 2019, they were now selling their own LiDAR system called Laser Bear Honeycomb at approx. $7500.
As implied, all of this works in conjunction with developments in AI and the computer chips on which the processes run from. Autonomous vehicle development therefore also has exposure on a section of industry working on computer chip design referred to as System on a Chip (SoC) design where the entire processing system is put on a single computer chip as opposed to multi-chip designs which are pricier, larger and use more energy. Research by Counterpoint forecasts that the use of autonomous vehicle SoC unit sales to be on the rise as vehicle autonomy increasingly improves but at different rates for the different autonomy levels as shown in figure 2.
Considerations for the Future
As intimated in the discussion, quite a few things need to come together from the technology front to get to a point where self driving cars become commonplace. Technology is taking great strides in edging closer but some key hurdles still need to be overcome. Also as we have seen with the case of the Level 3 package Audi A8 release, a lot of globally non-standardized red tape still exists with respect to the legal framework. We are therefore likely to see such issues e.g. insurance policies, develop some kind of standardization over time from precedence across different jurisdictions at differing pace.
One of the debatable aspects is whether technological innovation is enough to compensate for the complexity of real life driving that humans deal with daily such as sudden changes in road conditions. However, it is also recognized that humans are by in large the cause of most vehicle accidents. In South Africa for example, a paper by Vester & Fourie published on the South African Journal of Science puts the percentage of fatal road accidents due to human error at 80% and just as alarming, pedestrians are approx. 38% of fatalities. In the US, the National Highway Traffic Safety Administration (NHTSA) puts human error at an even higher 94 – 96%. Another factor to consider is the human driver contribution to traffic jams whether its in the form of poor on-road decision making, pure error or failure to respond to weather or road conditions . All this indicates the complexity autonomous driving would have to deal with when co-existing with human drivers also on the roads. Without a critical mass of the proportion of autonomous-to-human driven cars, it therefore seems technology would have to leap even further forward to compensate for this challenge.
Vester & Fourie: The good, the bad and the ugly of South African fatal road accidents
For these reasons, one can lean towards the belief that we are more likely to see autonomous vehicles have a better deployment success rate in self-contained ecosystems e.g. bounded zones of cities or smart cities where only autonomous vehicles are allowed. A further advantage of this would be that the concept of rules for the road ,traffic flow and scheduling would be able to synchronize with a central function, database or supercomputer. The network effect benefits of such a set up could be astronomic especially if also adopted in road based public transport.
Another important consideration regarding full vehicle automation relates to cyber security given that data is a fundamental element of delivering autonomous driving. Incidents of vehicle hacking have already been experienced though at levels not considered threateningly high. It however plasters some uncertainly as to how much damage more aggressive hacking of vehicle systems could be especially if they become more prevalent as expected. To this end, automakers such as Tesla and Chrysler have introduced ideas like hacking incentive programs where hackers who breach their systems can be rewarded by payouts in a counter-intuitive move to bring otherwise invisible vehicle data system vulnerability to light.
Final thoughts
It has been said that for humans born to this era of the world , technological progress feels like second nature to us – almost to the extent of being an entitlement. We have seen disruption and advancements in communication, entertainment, transport and various business models resulting in a disposition that expectation of disruption is built into our psyche. In addition to this, when we then look back, hindsight bias creates the illusion that these technological leaps where meant to be as we forget the incremental (and sometimes even lucky) nature it would have happened (see Futurewattage framing illusions article here). As such, we find that when thinking of what future vehicle technology holds, there is a tendency to overestimate what is possible in the long term future while on the other hand underestimating the short to medium term incremental changes. So while 2054 might not be Minority Report with self driving cars sliding along and slotting in and out of skyscrapers into lofty freeways , we are sure to see the compounded incremental effect of significant gains in vehicle technology. We are also sure to see this trickling down into safer more efficient vehicles in this imperial march forwards.
By Tare Kadzura
Other Useful Video Links
