Autonomous Vehicles: A Safer Road Ahead 

Recent data analyses underscore a promising trend: Autonomous vehicles (AVs) are showcasing remarkable safety records, a development that should warrant policymakers’ attention and ease public concern.

A Pennsylvania State University study has found that AVs have fewer crashes compared to conventional vehicles and less severe ones when they do occur. According to the research, AVs have been involved in 195 crashes over 4.62 million miles, which is 2.3 times fewer crashes than conventional vehicles per mile driven.

Not only do AVs crash less often; the crashes are far less severe. The vast majority of AV crashes (87.7 percent) resulted only in property damage. Meanwhile, 30 percent of crashes involving human drivers result in injuries, and 0.7 percent result in fatalities. 

Several factors contribute to AVs’ improved safety record. AVs constantly monitor their surroundings and can quickly apply brakes to reduce kinetic energy and crash impact. They follow traffic laws precisely and don’t make errors caused by distraction, intoxication, or fatigue. This enables AVs to avoid the riskiest driving maneuvers and situations that cause severe crashes. 

The researchers also found that AVs are particularly adept at avoiding rear-end, angled, and head-on crashes—the most dangerous collision types. More than 28 percent of conventional car crashes were front-end compared to just 7 percent of AV’s crashes. Rear-end crashes are more prevalent with AVs, but the researchers noted that they are mostly the result of a conventional car running into the back of an AV and that “the AV is rarely at fault.”

In a separate analysis, journalist Timothy B. Lee scrutinized every crash report filed in California by Waymo and Cruise, two frontrunners in the AV sector. His findings echoed the Pennsylvania State University study, emphasizing the minimal safety risk posed by the predominantly low-speed collisions involving AVs. Over the course of six million miles of driving, the two companies reported 102 crashes. That works out to one crash for every 60,000 miles, which is about five years of driving for a typical human motorist. Most accidents were low-speed collisions that did not pose a serious safety risk, and a “large majority appeared to be the fault of the other driver.”

The technology powering these safety features and driving capabilities is improving. Walter Isaacson wrote earlier this month about the radical shift Tesla is taking with its Full Self-Driving (FSD 12) software. Instead of being based on thousands of lines of code, the new system has taught itself how to drive by analyzing billions of frames of video of humans driving. It’s similar to how Large Language Models (LLMs) train themselves to generate answers by processing billions of words of human text. In this case, when confronted with a situation, the AV’s neural network selects a path or makes a decision based on what humans have done in thousands of similar situations. “It’s like ChatGPT, but for cars,” Dhaval Shroff, a member of the autopilot team said. “We process an enormous amount of data on how real human drivers acted in a complex driving situation, and then we train a computer’s neural network to mimic that.” Elon Musk live-streamed a demo of these complex capabilities, which featured the car navigating several situations it had never trained on before.

I had the chance to ride in a self-driving taxi Cruise ride in San Francisco and found it amazing. An app summoned the car to my location and, once I got it and buckled my seatbelt, it took off, navigating around stopped garbage trucks and avoiding jaywalking pedestrians. At no point did I feel unsafe or uncomfortable. In fact, it was a smoother ride than I’ve had using ride-sharing apps.  

These emerging data presents a compelling and promising case for AVs’ ability to significantly reduce traffic injuries and fatalities. It’s not perfect—for example, one Cruise got stuck in freshly poured concrete—but neither are human drivers. The small potential risks of this technology should be balanced against the advantages it offers, such as increased safety, more independence for some groups, and gains in productivity for individuals.

The most effective and quickest way to make AVs safer is actually to have them log more miles on the road, not fewer. More real-world driving experience for AVs means more data to refine their systems and improve safety. These feedback loops of encountering diverse scenarios are central to advancing AV technology and reducing risks. Regulators should adopt reasonable safeguards like crash reporting while taking care not to hinder progress with excessive caution. Some risks are inevitable with any new technology, but the benefits of accelerated AV learning could be enormous, including preventing many human driver errors and traffic fatalities. The path forward lies in enabling AVs to gain more on-road experience, not overregulating to restrict them.