On March 18, 2018, Elaine Herzberg pushed her bike across Mill Avenue in Tempe, Arizona. At ten o'clock at night, it would have been difficult for a driver to see someone cross the street, especially outside of a crosswalk. It wasn’t a driver, however, who hit the young woman. It was an autonomous vehicle.
Uber began testing its autonomous vehicles in Arizona back in 2016. After Herzberg’s death—the very first pedestrian death caused by an autonomous vehicle—Uber suspended its testing in the state. While proponents of the autonomous vehicle argue that these cars are much safer than conventional cars, which are subject to human error, critics point out that AI is flawed as well. And because computers cannot “think on their feet” unless specifically programmed to analyze and account for millions of possible scenarios, they are often less safe.
In this case, the autonomous car crashed due to software and sensor issues. The system was unable to distinguish Herzberg as a human rather than another vehicle, and because she was outside of the crosswalk, the system could not assign her an explicit goal (i.e. predict her path of travel). Unlike other companies—such as Waymo, Aurora, and Ford—Uber never referenced jaywalkers in its safety reports. In addition, the Uber’s sensors, which were not affected by darkness, were partially to blame, as they could not recognize Herzberg’s movement toward the left of the car.
According to the Society of Automotive Engineers, we aren’t even close to full autonomous function yet. And the more autonomous our vehicles become, the more dangerous.
Currently, most vehicles operate at an autonomous level of zero, which means the car possesses no autonomous features. At level one, cars can perform one autonomous task at a time, and at level two, cars can perform two or more autonomous tasks at a time. Most autonomous vehicles on the market are on levels one and two.
At level three, cars can drive without any human intervention, but only in certain conditions, and the driver must be able to take control of the wheel at any given moment. An example of a level three car is the Audi A8. Level four cars require no human intervention, but they, too, are restricted by certain conditions, such as speed or weather. Finally, level five cars are fully autonomous vehicles requiring no human intervention and with absolutely no restrictions. We have not yet developed level four and five autonomous vehicles, but with the safety concerns of the prior three vehicles, we might not reach these levels of autonomy.
Levels one through four, in which humans share a portion of responsibility with the vehicle, are the most dangerous levels because people will often believe they do not need to pay attention. This was the case with the Uber car that struck Herzberg, as the woman attending the car was looking at her phone rather than the road and did not take control of the vehicle to stop the collision. In addition to this, as people become more dependent on technology to perform tasks for them, they become less proficient in these tasks. When they are forced to take control, they may very well perform poorly and end up causing more collisions.
Dependence, human error, and software malfunctions are not the only things engineers worry about when it comes to designing autonomous vehicles. EMI also plays a part. EMI, or electromagnetic interference, occurs when an electronic or electrical device interferes with the function of another device. EMI has interfered with traditional cars in the past, as RF transmissions have reportedly triggered the release of air bags. Autonomous vehicles require a large number of electrical and electronic systems, which increases the potential for EMI. In addition to this, engineers will need to account for 5G networks. Traditional, level zero and one vehicles use AM and FM bands for radio and Bluetooth devices. Autonomous vehicle engineers will need to account for a greater range of higher frequencies and augment EMI shielding materials. If you have a product that needs to be tested for EMC compliance, contact Rhein Tech today. Our certified engineers and technicians have extensive knowledge of hardware design engineering, from product conception to final printed circuit board, with emphasis on EMC design principles. Sources: https://www.washingtonpost.com/technology/2019/11/11/what-self-driving-cars-cant-recognize-may-be-matter-life-death/, https://www.forbes.com/sites/cognitiveworld/2019/12/08/how-autonomous-vehicles-fit-into-our-ai-enabled-future/#2ca774bf5df9, http://blog.parker.com/electromagnetic-compatibility-in-next-gen-autonomous-vehicles