How Much Safer Are Autonomous Vehicles?
Human error accounts for up to 94% of current road collisions. Conventional thinking is that driverless cars will reduce those incidents by orders of magnitude.
The way to measure the safety of autonomous vehicles is miles per disengagement, which indicates how often a car needs a human driver to intervene. Waymo is currently the best at this metric with an annual average of 341 miles between disengagements.
Sensors
AVs need a whole host of sensors to see their surroundings. Sensors record data like speed, direction, and geolocation, then feed it to algorithms which help them interpret the car’s environment. Advanced sensors can identify obstacles that the human brain can’t, even down to the level of individual molecules (LIDAR).
Experts have long predicted that removing human error by allowing autonomous vehicles to do the driving would significantly reduce road accidents and save lives. This is a key selling point for these cars to an audience who remains skeptical of the technology.
Maps
It’s widely believed that human error is responsible for 85% of current road collisions. Removing humans from the equation through autonomous vehicles should therefore significantly reduce accidents.
One of the key factors in AV safety is high-definition maps. These provide a 3D representation of the vehicle’s environment and support key operational functions, such as localization, perception and navigation.
For example, Mercedes’s DRIVE PILOT, which is available in its new SAE Level 3 S-class, integrates HERE maps. This allows the vehicle to effectively “see around the corners” and drive safely within lanes.
The quality of HD maps is also important when a vehicle needs to parse the road – a task that requires the combined data from computer vision, radars and sonars. Without accurate mapping, AVs could run into other cars or become confused when they encounter road elements that don’t match their sensors’ readings.
Algorithms
The navigational algorithms that power autonomous vehicles interpret data from the vehicle’s sensors and turn it into safe driving decisions. They also use machine learning to adapt to changing road conditions and unexpected events, enhancing overall safety and reliability.
Quick Reflexes
Navigational algorithms enable autonomous vehicles to process information at lightning speed, making them much quicker than humans when responding to potential hazards. This speeds up the reaction time and reduces accidents caused by human error.
Consistent Compliance
Autonomous cars are programmed to obey traffic laws consistently, which minimizes accidents caused by reckless driving behaviors. The algorithms help them stick to speed limits, stop at red lights, yield to pedestrians, and make precise lane changes. They even know when to follow a car ahead of them at a four-way stop to ensure the vehicles clear the intersection in synchronized fashion. This makes it safer for other cars to pass through. They also avoid misunderstanding the subtle cues that humans read at stop signs and traffic signals, like whether it is safe to cross an intersection.
Communication
In addition to these systems, AVs must be able to communicate with each other and human drivers. They need to know who has the right of way at intersections, for example, and they need to tell each other when it’s safe to merge.
As you might have guessed, these systems can sometimes make mistakes. They can get confused by things like a brightly colored traffic cone or misread a speed limit sign. And they can even be fooled by fake signs that are placed on the road by hackers.
Despite these limitations, there’s been tremendous progress in autonomous vehicle safety. Last year, Waymo drove 440,000 miles without a single disengagement (which requires the safety driver to take control of the car). That’s comparable to one 200,000 mile average for human drivers in US cities. Obviously, more progress needs to be made before the public is fully ready for these vehicles. But one headline claiming that an autonomous car caused a collision will still trump 30,000 obituaries describing the lives lost because of human error on our roads.