How will autonomous cars handle complex hypotheticals?
A decade ago, the idea of a self-driving car only existed in the realm of science fiction movies and novels. Now, however, technology, artificial intelligence and machine learning have advanced to the degree that nearly all new cars have a host of autonomous safety features and several companies have real-world prototypes of self-driving cars and trucks being tested on the roads.
It is often reported that Tesla vehicles are shipped with all the necessary autonomous hardware, only needing a software upgrade to fully release the vehicle’s technological potential. When the future finally arrives, can it be truly estimated that self-driving vehicles will result in fewer collisions or more?
The problem of hypotheticals
The decision-making process can always be boiled down to its component parts – an “IF” query followed by the satisfactory “THEN” resolution. The more complex the hypothetical problem, of course, the more difficult it might be to reach the proper resolution. For example, IF traffic is stopped in front of you, THEN you must decelerate and stop. IF a pedestrian has stepped into a crosswalk, THEN you must stop to avoid a collision.
However, an autonomous vehicle might be forced into a scenario with cascading elements. Each time a driver gets behind the wheel, he or she is faced with situations influenced by both on-road and off-road factors. Pedestrians, animals, vehicles exiting a parking garage – these things are not in your path, but what IF they are?
Visual acuity and decision-making
While an autonomous vehicle will utilize countless sensors designed to recognize hazards and avoid danger, can every scenario be properly calculated? For example, are there certain situations where it might be impossible for a vehicle’s AI to properly determine the correct course of action?
- A disabled vehicle is pulled to the right shoulder with its hazard lights on. To safely navigate around the stopped car will require your moving car to cross the road’s center line. What parameters makes this a safe course of action? How far away does oncoming traffic have to be? Where are all the pedestrians?
- The autonomous car is navigating heavy traffic in a busy downtown area. The roads and the sidewalks are crowded. How does the car calculate the probability that a pedestrian will jaywalk directly in front of the vehicle? How does the car calculate the probability that a group of pedestrians at a crosswalk will wait until they are properly signaled before stepping into the street?
- For an intersection with a flashing yellow, semi-protected left-hand turn, how can the car calculate the odds of a safe turn if the oncoming traffic arrives around a corner or up a hill? What if there is another vehicle exiting a parking lot directly across the intersection? What are the odds of that car getting into the road and speeding through the intersection resulting in an unsafe left-hand turn?
Since these motor vehicle accident scenarios are identified as problems, it gives automakers the chance to uncover a solution. The hope is to develop sensors that can recognize hazards well in advance of the critical decision-making period. In the trolley-related thought experiment, if you could see the issue a mile in advance, it might be easier to reach a better resolution than simply deciding who dies.