Will autonomous cars coexist with human-driven cars at the same time?

According to a study by Prof. Amnon Shashua and Shilo Schwartz, when the autonomous car is implemented without drivers in all situations, it will have many economic and social consequences, such as the reduction of congestion on the roads * One of them will be the drastic decrease in the number of traffic accidents

Will autonomous cars coexist with human-driven cars at the same time? Prof. Tilapia amusement I think so. In an article he published together with Shay Shalow Schwartz It is written that when the autonomous car is implemented without drivers in all situations (level 5) it will have many economic and social consequences such as reducing congestion on the roads. One of them will be the drastic decrease in the number of traffic accidents.

But in the study published by the two, they write that autonomous transportation is not just a product, it is an industry - and to improve mobility, the car manufacturers, technology companies and government regulators need to develop a holistic model. According to Mobileye (Mobileye)-Intel (Intel), there is an innovative framework for this model. It focuses on two fundamental pillars: proven safety guarantees and economic scalability. Without a clear model for these two critical parameters, the investment in autonomous vehicle development will not pay off.

According to the two, "it is important for the automotive industry to cooperate in safety verification methodology and standards in partnership with global standardization bodies and regulators. The United States is among the leading countries in legislation in the field of autonomous vehicles. We are presenting a model for an autonomous vehicle system, the result of which is a drastic improvement in safety."

"A pre-determined set of rules must be established for rapid assessment and determination of responsibility for an accident in the event that autonomous cars are involved in a collision with cars driven by a person. The autonomous car will share the road with manned vehicles for decades, therefore in order for the autonomous cars to be mobile, they must operate like any other car - do not limit to a low speed or keep too much distance."

"If one of the manned cars makes an error and cuts off an autonomous car, the autonomous car has no way to avoid a collision. It is also impossible to prevent the autonomous car from being in such a state. The conclusion is that any autonomous car may be involved in situations that could lead to accidents, including mechanical failure and external forces."

What will happen when a collision occurs?

What will happen when a collision occurs? There will be an investigation, which could take months. Even if the human-driven vehicle was responsible, it may not be immediately clear. Public attention to the fact that this is an accident with an autonomous vehicle will be high.

Our solution, said the two, "is to establish clear rules according to a mathematical model. If the rules are set in advance, then the investigation can be very short based on facts, and it will be possible to determine who is responsible definitively. This will strengthen the public's trust in autonomous cars, when it will be clear that these failures are inevitable, and the risks and responsibilities for consumers, the automobile industry and the insurance industries must be clarified."

"Our approach regulates common sense in the investigation of responsibility for the accident. It describes how we can formalize the driving scenarios, defines concepts of priority and wins ahead and equations involving speed, distance, etc., and everything will be integrated within a formal mathematical model for determining faults".

"Since autonomous cars have an array of highly accurate sensors, there will be data available to assess the exact environment before and during the collision. In conjunction with the formal rules for determining faults. This data can be used quickly, to exclusively determine responsibility."

The fault model enables the creation of decision-making software (driving policy) that will prevent accidents caused by an autonomous car system.

"Autonomous cars have 360-degree vision and lightning-fast reaction times, and can analyze road conditions and available braking power, and are never distracted. In light of this, in addition to the official model for determining faults, autonomous car developers can design a system in which the software can assess any risk and not allow the vehicle to be sent a command that could lead to an accident while avoiding the intensive data verification process."

"Our system can reduce mortality from one death per million kilometers traveled to one death per billion kilometers. In terms of the United States, we will reach a death rate of 40 people per year compared to 40 who were killed on the roads of the United States in 2016."

However, they said, "Despite the algorithms, accidents caused by errors of the sensor system - that is, the information about the driving environment that is used as a basis for decision-making, or mechanical failure - can still occur. We propose to establish a fusion system that uses three independent engineering systems, each of which is based on a different technology - a camera, a high-resolution map, and radar and LIDAR sensors."

Absolute safety is impossible

Absolute safety is impossible. As long as the autonomous cars share the road with human drivers, we must ensure that the autonomous cars do not cause accidents. Essentially, "common sense" will be formulated through mathematical equations. These equations can be defined when the vehicle's ability to perform evasive maneuvers and steering and braking operations is known. Formalizing this model of assigning blame makes it possible to plan the decisions of the vehicles so that they respond to these parameters. Always evaluate the rules and never give a command that would put the vehicle at risk of causing a collision.

This way it will be possible to ensure that the autonomous cars will not deviate from the commands and will not cause an accident even if others do unexpected or reckless actions. The "safe mode" establishes a strict rule that autonomous cars will not execute commands outside of the set of caution commands. This ensures that the planning mode itself will never cause an accident.

Intel and Mobilya have already started working with BMW On a safe autonomous car platform.

In conclusion, the two write that "it must be clearly understood that without a scalable economic model, the true potential of autonomous cars will not reach the masses."

See more on the subject on the science website:

• The future of electric vehicles - and the meaning for all of us
• "Technology will make it possible to reduce traffic accidents by ninety percent"
• "People are willing to tolerate the death of 35 thousand people from vehicles driven by humans, but they will not be able to tolerate one death from a robotic vehicle"
• Automatic, not autonomous
• The computer also needs to pass a test