Mobileye’s CEO and founder Amnon Shashua has set out a mathematical model to ensure the safety of self-driving cars and also proposes a software design for autonomous cars which he believes enables them to be both safe and cheap enough for mass production.
Shashua presented his solution at the World Knowledge Forum in Seoul, South Korea, on Tuesday, based on an academic paper he wrote together with professors Shai Shalev-Shwartz and Shaked Shammah.
“There is little argument that machines will be better drivers than humans. Yet there is very real risk that self-driving vehicles will never realize their life-saving potential if we can’t agree on standards for safety,” Shashua and Shalev-Shwartz said on an open ad published with the paper. “We believe self-driving vehicles can and should be held to a standard of operational safety that is inordinately better than what we humans exhibit today. And the time to develop those standards is now.”
With a lack of safety standards and a lack of scalability — the option of mass-producing self driving cars — development of autonomous vehicles (AV) is “dangerously” moving along a path that might end in “great disappointment, after which further progress will come to a halt for many years to come,” the professors said in the paper.
To succeed, self-driving cars must operate in a normal way, wrote Shashua and Shalev-Shwartz in the paper – a layman’s version of which was also published by the two. This means they must not be restricted to very low speeds or to staying far away from other cars. It must also be assumed that complete avoidance of “every accident scenario is impossible,” they wrote.
The solution, they say, is to set out clear rules for fault in advance, based on a mathematical model. If the rules are predetermined, then any investigation into accidents can be “short and based on facts, and responsibility can be determined conclusively,” they wrote. This will help bolster public confidence in autonomous vehicles when such incidents happen, and help make liability risks clear for both the consumers and the automotive and insurance industries.
The professors set out a “commonsense” mathematical formula which includes a set of driving scenarios and how best to navigate them, with equations of optimal speed and distances, all of which combine to determine responsibility and fault in the case of an accident. Data collected by the sensors on the cars will also help give a fuller picture of responsibility.
Shashua and Shalev-Shwartz then use the fault-assigning mathematical model as the basis of decision-making, driving-policy software called Responsibility-Sensitive Safety (RSS). The software ensures that, from a planning and decision-making perspective, the AV system will not issue a command that would lead to an accident. The system would be set up to avoid cut-in maneuvers, for example, and be taught how to be careful in situations in which sight is limited; it would also help determine route priorities, how to manage a two-way traffic situation and how to handle traffic lights, for example.
The professors also propose AVs be equipped with three independently engineered systems that rely on different technologies: a camera, high-definition maps, and radar and lidar systems that can be combined to support a “minuscule error rate.”
Intel Corp. acquired Mobileye in March for a whopping $15.3 billion.
“Intel and Mobileye have already begun work with BMW on a nonexclusive platform for AVs that follows these concepts,” Shashua said in the layman’s paper. “It is an example of an inclusive relationship of partners that each share a vision to put safety first. We also clearly understand that without an economic model that is truly scalable, the real potential of AVs will never be experienced by the masses.”
In his talk, Shashua also called upon regulators and standard bodies to come forward with additional scenarios that need to be evaluated, and to “collaboratively construct standards that definitively assign accident fault.”
All the rules and regulations today are framed around the idea of a driver in control of the car, he said, and thus new parameters are needed for autonomous vehicles.