The Potential Winners and Losers Due to Relaxing Autonomous Driving Safety

It often seems as though every discussion concerning the automotive market is punctuated by the word “safety.” Every enabling technology supplier targeting the automotive market in recent years has come to respect the importance of “automotive-grade design” and most have learned that the key to achieving growth in the automotive market has been to tie to their technology to some type of automotive safety application.

While safety touches on every automotive domain, it has defined the pace of autonomous driving in recent years, as the task of improving Autonomous Vehicle (AV) safety (relative to manually driven vehicles) and demonstrating this level of safety within a cost envelope that is appropriate for the passenger vehicle market has brought stagnation to the unsupervised autonomous driving market.

It is, therefore, nothing short of extraordinary that many in the automotive industry should find themselves contemplating the ideal strategic response to a potential shift in this safety-centric approach to autonomous driving, with many automakers and suppliers anticipating that the new U.S. administration will look to accelerate AV development and close a perceived innovation gap with China by introducing a new policy environment that de-emphasizes safety and demonstration of safety in favor of scale.

At the time of writing, the market can only speculate as to what this new policy environment might look like, but based on the assumption that the primary motivating factor behind any change will be to close the perceived innovation gap with China and accelerate deployment, the primary lever at the disposal of policy makers would be to reduce the threshold for the safety of autonomous driving.

There is currently no universally accepted standard for defining the safety of an AV system; however, most industry commentators and vendors within the market tend to use Mean Time Between Failure (MTBF) or driver intervention as a proxy for how safe an AV is. Setting aside the potential weaknesses of such a measure, in general, this measure allows for relatively easy comparison between the safety of manually driven vehicles (deriving from accident statistics how many miles driven/hours of driving occur between accidents) and the safety of AV products or prototypes (measuring how frequently consumers or specialist safety drivers intervene to prevent a collision).

While there is no definite figure held by all in the market or even communicated by regulatory or governmental bodies, most vendors target a significant improvement in the MTBF compared to human operation, such as 106 or 107 hours of driving—an order of magnitude improvement over manual driving. This is partly due to the fact that the MTBF statistics for manually driven vehicles presumably reflect factors such as drunk driving, distracted driving, and other human factors that should not even be relevant to autonomous systems.

Delivering on this “order of magnitude” safety improvement has mandated the development of a series of independent and redundant technologies, which when fused together amount to a level of reliability in perception and path planning commensurate with high MTBF. Specifically, sensors such as Light Detection and Ranging (LiDAR) and, to a lesser extent, imaging radar, as well as the use of redundant algorithms for perception, environment modeling, and path planning, require an increase in the embedded compute resources of the vehicle.

Therefore, if a policy framework were to emerge in the United States requiring an autonomous MTBF that is close to or only moderately better than human MTBF, it would diminish the relevance and market potential for tertiary sensors, redundant algorithms, and, to a lesser extent, high-performance compute, as some compute headroom would be devoted to supporting these redundant algorithms. LiDAR products would be particularly impacted, especially those currently targeted for Level 3 autonomous driving in the United States and Western Europe.

Beyond the impact on redundant components, a shift in the automotive safety culture toward lower MTBF would have a disproportionate impact on AV platforms on a system level, depending on their approach to sensor fusion. Some developers have favored an approach of low-level sensor fusion, combining raw data from multiple sensors (including tertiary sensors like LiDAR and imaging radar) directly and before additional processing. If LiDAR and imaging radar lose their relevance as redundant sensors targeting high MTBF, AV platform developers that have leveraged low-level fusion will face a greater engineering effort in redeveloping their sensor fusion algorithms to accommodate the new optimum sensor configuration. In contrast, some AV platform developers have preferred high-level fusion, processing the inputs from each separate sensor modality independently, before combining them in an effort to leverage the unique qualities of each sensor type to build a better understanding of the environment. Should the expected safety culture shift take place, vendors that have pursued high-level sensor fusion will be in a better position to fall back on the new optimal sensor configuration, which will likely be vision-centric, perhaps leveraging some small amount of conventional automotive radar.

There are already AV products in the market that rely solely on a combination of camera and radar sensors, often fuse at a high level, but have a relatively low MTBF, and therefore require a human driver to provide ongoing supervision to build a stringent safety case—often referred to as Level 2+ autonomous driving. If a new policy environment were to emerge that accepted a low MTBF for unsupervised autonomous driving, then it would effectively allow for today's Level 2+/supervised technology stacks to power unsupervised autonomous driving experiences. Therefore, vendors with an existing foothold in the Level 2+ market would be best placed to take advantage of a shift in the autonomous driving safety culture.

Of course, the discussion, so far, has only considered “winners and losers” from the perspective of different technology developers and component suppliers, and many in the industry are concerned that a shift in the safety culture could result in the general public being the biggest “losers” overall. Indeed, many suspect that this policy shift could only come as a result of the perceived influence of Elon Musk on the current U.S. administration, with Tesla’s Chief Executive Officer (CEO) having made clear his rejection of many of the technologies associated with redundancy—LiDAR, High-Definition (HD) maps, and even radar to certain degree.

Therefore, if high MTBF ceases to be a given for unsupervised autonomous driving in the United States, it will likely become a differentiator. U.S. Original Equipment Manufacturers (OEMs) are aware of how quickly GM’s Cruise operation disintegrated and are highly risk averse and conscious of how AV accidents can impact their brand reputation.

Meanwhile, the ideal strategic shift for suppliers of redundant technologies would be to generate new products pivoted away from delivering redundancy and toward supporting greater consumer value. To an extent, this is already visible in the Chinese market, where less performant LiDAR has been deployed on Level 2+ vehicles in order to improve the consumer experience, rather than to help build a safety case. Similarly, there is a clear preference in China for sparse array imaging radar, as the consumer experience improvements delivered by higher resolution are preferred over the improved reliability that comes from denser arrays. A scalable set of technologies would allow these suppliers to deliver both higher MTBF for risk-averse OEMs, and more apparent consumer value for OEMs prioritizing deployment.