Autonomous Autos – The heavy hand.

NHTSA takes aim at Tesla.

• The NHTSA (US vehicle safety authority) has ordered that all crashes that involve vehicles that are equipped with Level 2 and above autonomous driving systems must be reported in what can only be a warning shot across Tesla’s bows.
• A level 2 autonomous driving system is basically adaptive cruise control with lane-centering capability which is now widely available on most vehicles that are sold at the upper end of the market.
• Reports must be made whether or not the system was engaged and a report must be made every month even if no crashes occurred.
• This is going to cause all sorts of difficulties because this requirement falls on the manufacturer of the vehicle and not the vehicle owner.
• Almost all OEMs lose track of the vehicles that they make as soon as they leave the forecourt and so complying with this regulation is going to be extremely difficult or impossible.
• However, I think that this is aimed mostly at Tesla which knows where all of its vehicles all of the time and exactly what they are doing.
• Tesla is also viewed as the cowboy of the automotive industry by its peers as it routinely allows its customers to test pre-release software as long as they assume complete responsibility for their actions.
• Unfortunately, some of these actions have been ill-advised and as a result, some lives have been lost.
• This is what the NHTSA is aiming to prevent, and this also signals the beginning of much more rigorous oversight of the autonomous driving industry.
• This is where some real problems may begin as many players use deep neural networks for perceiving the environment and these systems constantly make mistakes.
• Furthermore, these systems are black boxes such that when a mistake is made there is no way of telling why it made this mistake.
• This excuse will not sit very well with the regulators which is why I have long suspected that autonomous driving systems with a large deep learning component will struggle to meet safety requirements.
• This new requirement by the NHTSA is a sign that this is exactly where the regulation of this industry is heading.
• RFM research indicates that the architecture with the best chance of success is one where deep learning is used in small specific tasks such as recognising pedestrians, traffic lights, cyclists, and so on while the decision making is carried out in software.
• This has the advantage of being both explainable which will keep the regulators happy and potentially far more energy efficient which will lead to better range from batteries.
• There is still a long way to go and the regulator has signaled that a more cautious approach to testing is required which will slow things up further.
• RFM maintains its target of 2028 for commercial self-driving to hit the market meaning that revenues are still quite far away.
• The net result is that consolidation is going to continue which combined with its persistent overvaluation makes it a sector still to avoid.