Autonomous driving ain’t easy. While Tesla CEO Elon Musk promised back in 2016 that its cars would support fully-autonomous cross-country driving by the end of the following year, he admitted last week that it’s “extremely difficult to achieve a general solution for self-driving that works well everywhere.”

Tesla’s hardly alone in terms of autonomous driving projects that have hit roadblocks: Uber laid off most of its autonomous driving unit earlier this year. Waymo, another industry leader, was also the subject of an August report in The Information which said that its cars struggle to turn left on faster roads and fail to interpret basic street features.

Taken together, it all begs the question of how so many smart predictions got ahead of themselves in terms of what it would really take to bring autonomous driving to the road. Part of the problem may have to do with underestimating how differently we all drive.

“I agree with Musk that self driving in its current state does not work well everywhere,” Shaoshan Liu, co-founder and chairman of autonomous robotics firm PerceptIn, tells Inverse.

Liu identifies two big roadblocks for mass adoption. The first is that autonomous cars are incredibly expensive right now, costing between $500,000 and $1 million to build, and most run simply for demonstration purposes. Tesla is promising full autonomy for its complete range that starts at $35,000, conditional on two software unlocks that can cost as little as $8,000 together, but this has yet to reach consumers.

The second impediment is that, despite the high cost, autonomous cars still don’t work very well. They struggle in most everyday situations, making them impractical for wider use. Despite the report, Waymo is still widely considered among the frontrunners in the autonomy race, with plans to launch its autonomous taxi service in Arizona later this year.

Waymo Google and Lyft are teaming up to take down Uber on self driving cars
Even Waymo struggles with driving.

Tesla has faced similar issues. Autopilot, its driver assistance feature that’s designed as a stepping stone to full autonomy, was meant to include a “Navigate on Autopilot function that could steer a car off the highway at the correct exit, but Musk held the feature back at the last minute for “a few more weeks of validation.”

Some problems can even have devastating consequences. An autonomous Uber killed 49-year-old Elaine Herzberg in March, becoming the first person ever killed by a self-driving vehicle. The incident caused the company to immediately put the brakes on its autonomy program.

Experts have suggested a number of solutions to utilize the technology in a safer but still effective manner. Andrew Ng, whose venture fund invests in autonomous car firm Drive.AI, told Bloomberg that a good solution could be completely segregating autonomous cars from pedestrians. Gary Marcus, a professor at New York University, dismissed Ng’s idea as “just redefining the goalposts to make the job easier.”

“In order for this technology to progress, we have to begin by adopting autonomous driving technologies in low-speed scenarios, such as university campuses, industrial parks etc,” Liu says. “Low-speed environments make it easier and much more affordable to ensure safety, thus allowing immediate deployment to deliver the benefits of autonomous driving. Then, as technology advances and experience is gained, it will be possible to consider gradual upgrading to higher speeds, ultimately exhibiting vehicle performance equal that of a human driver in any driving scenario.”

Accidents will happen though, and the world needs to prepare for that. Shashi Verma, chief technology officer at the Transport for London government body, said at a November 2017 event that “the fundamental problem that the automotive industry would have to solve is [while] the world is perfectly happy killing half a million people on the roads every year…the first robot that kills a person will be a news story.”

Experts are still deeply divided on how cars should respond in such a situation. A study published this week in Risk Analysis: An International Journal entitled “How should autonomous cars drive? A preference for defaults in moral judgments under risk and uncertainty,” conducted a survey and found respondents prefer autonomous cars to stay in their lane and perform an emergency stop rather than take more drastic action. A representative from Mercedes-Benz, however, said in 2016 that the car should prioritize saving the driver over bystanders, as it’s the variable the car can best control. These two philosophies could conflict in the real world.

A Mercedes-Benz driving itself.
A Mercedes-Benz driving itself.

So when will autonomous driving really reach the road? Musk has suggested cross-country trips could be achieved with an alpha version of the next software update. Waymo plans to expand out its limited Arizona taxi trials to more consumers gradually over time. Computer chip developer Arm, however, claimed last month that real production of full autonomous vehicles for consumers could come as late as 2027, around 10 years after Musk’s original deadline.

The dream of going to sleep and having your car whisk you to another place could remain a dream for most of the coming decade.

Photos via Waymo, Mercedes-Benz/YouTube, Tesla