In the section self-driving vehicles, the report states that sensing algorithms will achieve super-human performance for capabilities to meet the requirement of driving in the near future. This statement is based on the situation that the current automated perception is already near or at human-level performance. However, it is not suitable to equal the sensing algorithms to human being just based on several performances now.
It is true that the Google self-driving car could use some sensors to pick up on details in the environment to help it identify a more precise location. It could recognize the STOP sign on the road and check the situation before moving forward. However, two challenging things for this. First, if there were four STOP signs at a crossroad and four self-driving cars reach their separate STOP sign at almost the same time. How could the car decide whether to go or yield? This might be solved by establishing the communication between different cars. In fact, the world is more complicated than the sensing algorithms. For instance, the electrical power system was broken by the strong wind in Detroit in March, 2017. The traffic lights did not work at the crossroad and there were no STOP sign at the crossroad. For such kind of situation, all human drivers of the north-south side would yield the drivers of the east-west side first, and then go across the road as if there was a default and invisible STOP sign at the crossroad. It seems to be hard for self-driving cars to make such kind of behaviors based on the sensing algorithms. Second, one assumes that a pedestrian stood close to a bus stop who might want to go across the road or wait for the bus. It is very difficult for the car to make a decision like a human being to beckon to the pedestrian to let she/he go first. Thus, it is a little early to say the sensing algorithms could achieve super-human performance in the near future.
Again, self-driving cars use sensors to check the surrounding environment, collect the information, process the information, and make a final decision. Imagine two situations when the car stops in a valley. The first situation is there might be some rocks of different sizes falling off, and the second situation is some birds might fly to the car and rest on the car roof. The answer to whether the falling rock might damage the car depends on the size and falling off speed of the rock. In general, birds will rarely cause damages to the car in the second situation. Then how the so called super-human sensing algorithms would be developed to tell the difference between the two situations and to make a decision if the car should be started and moved away. People usually make decisions based on the complicated neutral work which has not been understood clearly up to now. Thus, it seems there will be a long way to equip the sensing algorithms with super-human performance.
In sum, self-driving car is an encouraging application of artificial intelligence which could make some human-like decisions and behaviors based on the algorithm calculation. However, it is still an obstacle to extend those algorithms to be universal and to determine the situations as human beings. Thus, there is still a long way to go to develop sensing algorithms of super-human performance.
