Mobileye Responsibility-Sensitive Safety (RSS)

Responsibility-Sensitive Safety (RSS) is the mathematical model of unmanned vehicles (AV) offered by Mobileye, Intel subsidiary company.

In general the RSS model is not the system allowing to avoid responsibility, but represents the innovation model which, as envisioned by Mobileye, will allow to achieve high standards of safety of operation of unmanned vehicles.

That the RSS model does not allow

• RSS does not allow unmanned vehicles to make such decisions which would lead to accident.
  • On the other hand, the RSS model nevertheless gives the chance to unmanned vehicles to execute maneuvers in abuse of regulations, for example, to cross a solid line of a marking to avoid collision or to go round the car parked by second nearby in order to avoid danger. However the RSS model does not allow to execute to the unmanned vehicle any unconstrained maneuvers which in itself can lead to separate accident.

• The RSS model does not allow the unmanned vehicle to minimize accident effects due to commission of other accident, let and with smaller effects.
  • In other words, trying to avoid collision with the car managed by the driver person, the RSS model allows the unmanned vehicle to make any maneuvers (even in traffic offense) if these maneuvers do not lead to other accident. This restriction quite appropriate as judgments of scales of accident are subjective and any hidden, critically significant factors can not consider, for example, the child on back seat of the car with which accident, apparently, would allow to minimize effects of the first.

• In the RSS model also the reasoning according to which the right of the primary movement can use only when it is provided is formalized rukovodstvuyemy by common sense.
  • Example: the car crosses the intersection. Green light grants to the car the right of the primary movement on the intersection, but that if the intersection was occupied with other vehicle (for example, passed on red light). In this case the RSS model does not grant to the unmanned vehicle the rights to collide with the car, arisen on the road – differently in this case the unmanned vehicle, according to the RSS model, will be guilty of accident.

Nevertheless, if the public finds it necessary to save an opportunity to minimize effects of one accident at the expense of another, then under certain conditions such algorithm can be added to a formula of RSS, having saved at the same time the principle of "succession of fault of responsibility" according to which responsibility for all incidents will be conferred on that participant of traffic because of which all chain began. We decided not to use this opportunity in our model yet, but it is quite implemented — professor Amnon Shashua, the CEO and CTO Mobileye, the senior vice president of Intel explained.

2017: Development

As it became known in November, 2017, the Mobileye company developed and continues to develop a mathematical model of Responsibility-Sensitive-Safety (RSS) unmanned vehicles which shows need of formation of standards of determination of responsibility in case of accident and also standards of safety of unmanned vehicles.

When developing the Responsibility-Sensitive Safety model in the company calculated that it will serve as a peculiar catalyst for a cross-industry discussion in which different industrial groups, car makers and governing bodies will participate.

As the RSS model formalizes dilemmas when driving

As professor Amnon Shashua explained, to the CEO and CTO Mobileye, the senior vice president of Intel, RSS formalizes common sense of human judgments in the most different road situations. The model sets accurate determinations of the fact that it means safe driving and that, on the contrary, belongs to dangerous driving. When the person drives the car, interpretation of responsibility in case of accidents or other incidents can vary quite widely. It is not important whether there was an accident by mistake of the driver or not, responsibility anyway is established proceeding from insufficiently complete information and other factors which are interpreted a post factum. In case of autonomous driving, these determinations can be formalized and presented in the mathematical form.

Machines have extremely exact information on surrounding objects, always register time of the reaction and brake efforts, they never distract from a road situation and do not know fatigue. Therefore when the computer drives the car, it is not necessary to interpret its actions a post factum. Instead it is possible to program it so that it followed a certain algorithm or a certain standard circuit (determined pattern). Everything that for this purpose it is necessary — to have an opportunity to formalize this scheme, Shashua explained.

In essence, the RSS model is created for formalization of those dilemmas which people face when driving and also for their premises in a specific context. In particular, the model allows:

• formalize concepts of a safe distance and safe distance at maneuvers of evolution or a vklinivaniya in a flow,
• define the participants of traffic performing evolutions timely to take necessary measures and to observe a safe distance,
• formalize the right of primary journey,
• give definition of safe driving in case of limited perception (for example when participants of traffic can be hidden behind buildings or the parked cars and suddenly to appear on the road), and many other things.

In general the RSS model tries to build a formal basis which would set all aspects of human judgments in a driving context for the purpose of creation of a peculiar "printing of security" (seal of safety), formal security accreditation of unmanned vehicles.

We cannot be limited to use of the RSS model only in our technologies of autonomous driving. To provide full-fledged security accreditations, we will demand public acceptance and formation of transparent standards of integration of human judgments in decision making by unmanned vehicles. We are convinced that we will be able to make traffic one thousand times safer in terms of the accidents caused properly by the designed unmanned vehicles in comparison with the accidents provoked by people — Amnon Shashua said.

Robotics