The tragic death on May 7 of a Tesla owner, whose car crashed into a tractor-trailer while in Autopilot mode, demands expanded debate on the subject of readiness and safety of autonomous driving. The topic involves a complex interplay of technology evolution, human behaviour and a driver’s ability to manage ambiguities. Contrary to some chatter that has erupted, there can be no simple binary conclusions. We need to be prepared to deal with many shades of grey.
AutonomyThe U.S. Department of Transportation has defined five levels of autonomous driving capability. They range from Level 0, represented by conventional autos of today, to Level 5 where future cars, like the Google Pod-Car, will have no opportunity for driver involvement. Most experts reckon that the Tesla Autopilot sits somewhere between Level 2 and Level 3, which means the car is capable of some degree of autonomous driving but has many limitations. Such vehicles can confront many traffic situations, where the system’s capabilities will be overwhelmed. Drivers require a deeper and more subtle understanding of what each system can manage and where driver involvement is necessary. Over time, system capabilities will expand, incorporating technologies like Vehicle-to-Vehicle communication and improving safety margins.
Human involvementThe use of an Autopilot function requires constant monitoring of the environment by the driver, who must remain ready to intervene and override the system when necessary. Human factors specialists will caution that such ambiguity is difficult to manage when one deals with a large population of drivers. These systems seek to offer the benefits of lower driving stress, which implies allowing the driver to be lulled into a sense of relaxation, only to be suddenly alerted to a situation that the autonomous system cannot manage. Such requirements place a high tax on a human’s ability to quickly grasp a situation and take meaningful evasive action. Automakers need to think twice before they allow customers to participate in “beta-testing” such safety-critical systems on public roads, even after suitable disclaimers.
CommunicationAutomakers must be concerned about how system capabilities are communicated to customers. The subtle distinctions between “autonomous driving” and” driver assist functions” are important and convey very different sets of expectations to customers. In a short period of time, many technologies have gone from being dismissed as “science fiction” to being the latest “must-have feature.” The rapidity of these transformations, accelerated by a world of hyperbole and advertisement, often allows insufficient time for technology maturation. For life-critical systems, as opposed to shopping apps on a mobile device, it cannot be sufficient to merely tell the customer to download the latest upgrade.
Regulatory oversightUnlike aviation, where future technologies are subjected to rigorous evaluation prior to certification, regulations in autos often lag technological advances. In this era of explosive growth in electronics, connectivity and machine intelligence, automotive regulators have struggled to keep up. Regulations need to strike a balance between encouraging innovation and protecting customers from “prototypes” that represent work in progress. While many U.S. states have taken positions on where and how fully autonomous cars may be used or tested, they have been unable to address use and deployment of partial autonomous capability including parking assist, autonomous valet and autonomous lane tracking. This area needs urgent attention even as the U.S. oversight body (NHTSA) has opened an investigation over the current accident. A former NHTSA attorney has confirmed that this investigation will cause the administration to venture into new territory – both for the nature of the review and its focus on complex operating software.
LiabilityMany industry experts believe that deployment of autonomy will be paced by evolving a legal understanding of liability and accountability. We can conceive of situations where accountability may be apportioned to the automaker, driver, owner, car-share user and even a third party who may cause an accident. When the car is in autonomous mode, it will require considerable amount of data and understanding of software to apportion liability. A brand new domain of the ethics of decisions made while in autonomous mode remains poorly studied.
A tragic loss of life has sharpened our awareness of the tasks ahead. Such complex situations cannot be managed with binary views on whether a technology is right or not. Evolution of technologies and human assimilation of new capabilities occur incrementally, leaving us to deal with periods of ambiguity and many shades of grey.
Managing these ambiguities, even as we sustain technological innovation and progress, is vital.
The writers are authors of ‘Changing the Machine’, a forthcoming book that deals with the Future of Automobility.