Unpacking the Safety Concerns Surrounding Tesla’s ‘Full Self-Driving’ System

DETROIT (AP) — In the last four months, William Stein, a tech analyst at Truist Securities, has taken Elon Musk up on his offer to test out Tesla’s much-talked-about “Full Self-Driving” system three times.

Tesla claims that their cars with this technology can go from one location to another with minimal human involvement. However, each time Stein drove one of these vehicles, he observed that it made unsafe or illegal moves. During his latest test drive this month, he said the experience was so alarming that it left his 16-year-old son feeling “terrified.”

Stein’s experiences—alongside a fatal crash involving a Tesla using Full Self-Driving technology in Seattle last April—have raised eyebrows among federal regulators. These authorities have been investigating Tesla’s automated driving systems for over two years due to numerous crashes that have sparked safety worries.

These ongoing issues have made those keeping an eye on autonomous vehicles more doubtful about Tesla’s ability to safely roll out its self-driving system on a larger scale. Stein himself is skeptical that Tesla will meet Musk’s prediction of launching a fleet of self-driving robotaxis by next year.

These recent incidents come at a crucial time for Tesla. Musk has told investors that by the end of this year, or possibly next, the Full Self-Driving system could operate more safely than human drivers.

In less than two months, the company plans to introduce a vehicle specifically designed for ride-sharing as a robotaxi. For Tesla to deploy these robotaxis, Musk has stated that they need to demonstrate to regulators that the system can drive more safely than people. Under federal guidelines, these Teslas must comply with national safety standards.

While Musk has shared data showing the distance each Tesla travels before a crash occurs, this information pertains only to the simpler Autopilot system. Safety experts criticize this data as misleading, noting that it only accounts for serious accidents where airbags were deployed, overlooking how often drivers had to take control to prevent collisions.

Approximately 500,000 Tesla owners—over 20% of all Teslas currently on the road—are using the Full Self-Driving feature, most having paid $8,000 or more for the privilege.

However, the company reminds users that cars equipped with this feature still cannot drive themselves entirely. Drivers must always be prepared to intervene if necessary. Tesla also monitors driver behavior and can restrict access to the Full Self-Driving feature if they fail to pay adequate attention. Recently, they have begun labeling the system as “Full Self-Driving (Supervised).”

Musk has previously admitted that some of his earlier forecasts about autonomous driving were overly ambitious. In 2019, he claimed a fleet of fully autonomous vehicles would be on the roads by the end of 2020. Now, five years later, many industry observers question whether this will ever happen broadly across the U.S.

Michael Brooks, executive director of the Center for Auto Safety, expressed his doubt, stating, “It’s not even close, and it’s not going to be next year.”

The vehicle Stein tested was a Tesla Model 3—the company’s most affordable option—equipped with the latest Full Self-Driving software. Musk claims the software now utilizes artificial intelligence to better control the car’s steering and acceleration.

Stein described the ride as smooth and more human-like compared to previous versions. However, during a short drive of less than ten miles, the car made a left turn while running a red light. “That was stunning,” Stein remarked. Although the road was relatively clear at the time, he didn’t take control immediately, but he later had to intervene when the car began to straddle two lanes on a busy parkway.

In his feedback to investors, Stein noted that the latest Full Self-Driving version does not “solve autonomy” as Musk had claimed, nor does it appear to be nearing the capabilities needed for robotaxis. He recounted similar unsettling experiences during earlier test drives in April and July.

Tesla has not responded to requests for comment on Stein’s observations.

Stein believes Tesla may eventually profit from its driving technology, but he isn’t optimistic about the immediate availability of a driverless robotaxi with a passenger in the back seat. He foresees significant delays and restrictions on where these vehicles can operate.

He pointed out that there is often a wide gap between Musk’s statements and the likely outcomes.

Some Tesla enthusiasts have shared videos on social media showing their cars driving themselves without human intervention. However, these videos do not provide insight into the system’s performance over time, and others have highlighted instances of the cars behaving dangerously.

Alain Kornhauser, who leads autonomous vehicle research at Princeton University, shared his experience of borrowing a Tesla for two weeks. He found that it usually recognized pedestrians and detected other vehicles, but he had to take control when the car made sudden, concerning maneuvers. He cautions that Full Self-Driving is not ready for complete autonomy in all environments.

“This thing,” he stated, “is not at a point where it can go anywhere.”

Kornhauser believes that the system might work autonomously in smaller neighborhoods where the streets are well-mapped. He questions why Musk doesn’t begin by offering smaller-scale rides in those more manageable areas, stating, “People could really use the mobility that this could provide.”

Experts have long warned that Tesla’s reliance on cameras and computer vision can fail to accurately recognize objects, especially in poor weather or low-light conditions. Most other companies developing robotaxi technology—like Waymo and Cruise from Alphabet and General Motors—use a mix of cameras, radar, and laser sensors for better accuracy.

“If you can’t see the world correctly, you can’t effectively navigate or move safely,” emphasized Missy Cummings, a professor at George Mason University. “Cars can’t do it with vision alone,” she clarified.

Cummings added that even companies utilizing laser and radar technology still struggle to achieve reliable driving outcomes, raising safety concerns for Waymo and Cruise as well. (Representatives from both companies declined to comment on these issues.)

Phil Koopman, a professor at Carnegie Mellon University focused on autonomous vehicle safety, predicts it will take many years before self-driving cars relying solely on artificial intelligence can respond effectively to all possible real-world scenarios.

“Machine learning lacks common sense; it learns in a very narrow manner based on numerous examples,” Koopman noted. “If the computer encounters a situation it hasn’t been programmed for, it risks crashing.”

In April, a Tesla using Full Self-Driving hit and killed a motorcyclist in Snohomish County, Washington. The driver, who has not been charged, reportedly was using Full Self-Driving while distracted by his phone when the collision occurred. The motorcyclist died at the scene.

Authorities are currently reviewing information regarding the crash from both Tesla and law enforcement officials. The National Highway Traffic Safety Administration (NHTSA) is also aware of Stein’s experiences with Full Self-Driving.

The NHTSA is investigating whether a Tesla recall implemented early this year, aimed at enhancing the automated driver monitoring system, was effective. Additionally, they urged Tesla to recall its Full Self-Driving feature earlier in 2023 due to concerns that, “in certain rare circumstances,” it may ignore some traffic laws, increasing the likelihood of a crash. (The agency declined to comment on whether the evaluation of the recall’s effectiveness has been completed.)

As Tesla’s electric vehicle sales have declined in recent months despite price cuts, Musk has advised investors to consider the company as more than just an automobile manufacturer, instead framing it as a robotics and artificial intelligence enterprise. Nonetheless, Tesla has been developing its Full Self-Driving technology since at least 2015.

“I suggest that anyone who doubts Tesla’s ability to achieve vehicle autonomy should not hold Tesla stock,” Musk stated during a recent earnings call.

However, Stein cautioned investors that they should independently assess whether Tesla’s Full Self-Driving, which is “the most established project of the company, currently generating revenue, and being tested in the real world,” actually functions as advertised.