So which one is the right approach? We may have the answer in front of us in the form of the well-established process that is used to teach humans how to drive and grant them a driving license. Upon passing a written test proving basic theoretical knowledge of the rules of the road, applicants are provided with a learnerâ€™s permit, allowing them to drive under certain conditions, usually involving an experienced driver sitting next to them, and restrictions on the time of day, the types of roads they are allowed to drive on, and the kind of vehicles they may operate. This is typically augmented by self- or instructor-induced limitations, by which new drivers typically acquire basic skills in secluded locations such as parking lots or back streets, gradually moving on to more challenging scenarios. In each of these situations the novice driver is expected to be fully responsible for driving Warnings or overrides the car; rather than performing only one or two functions, or may happen, but these are relying by design on the instructors to avoid an obstacle, keep perceived as driver failures the speed limit, or stop for a traffic light. Warnings or overrides rather than a means to (e.g., the experienced driver-passenger seizing the controls) may happen, but these are perceived as driver failures rather than a mitigate difficult situations. preferred means to mitigate difficult situations. So, why not follow a similar process for fully autonomous cars? Let Level 4 automation systems have complete authority on the vehicle, but restrict the conditions under which they are able to operate, possibly mandating the presence of a professional safety driver onboard, and appropriate workload conditions for the safety driver. Once the vehicle can demonstrate increasing level of competency (possibly by a combination of real and simulated driving records, using naturalistic or synthetic data), to the satisfaction of the relevant authorities, reduce the restrictions until the point at which the automation is deemed safe to operate at the desired level.
Published on Nov 18, 2016
Annual magazine review of MIT Aeronautics and Astronautics Department research and educational initiatives.