Tesla has been under intense criticism after a fatal accident in May 2016, when Tesla driver Joshua Brown’s Model S, in Autopilot mode, crashed into a tractor-trailer truck. Neither Brown nor Autopilot applied the brakes.

Since then, there’s been much debate about Autopilot, with many wondering if Tesla developed a system that drivers should not always trust.

In fact, drivers should not completely trust the system–it’s still in beta mode, and Tesla’s guidelines for use emphasize that the autonomous feature should only be used for highway driving, and only when the driver is fully paying attention.

On Sunday, Elon Musk announced an upgrade to Autopilot, which should be rolled out in the next week or two.

The move is “a definite response to the fatality,” said Jeffrey Miller, IEEE member and associate professor of engineering at the University of Southern California. In fact, Musk himself said that the new software update would have prevented Brown’s death.

So what exactly is involved in Autopilot 8.0? Here are some of the key takeaways.

Emphasis on radar

To detect obstacles in the car’s path, Tesla is focusing on radar. Its current vehicles do not have LiDAR, the laser system used in many autonomous systems like those developed by Google and Ford. “Tesla’s Autopilot is apparently using radar in a way that most other driving systems use LiDAR,” said Bryant Walker Smith, professor at the University of South Carolina and autonomous car expert, “as the primary perception technology for understanding the world through which the vehicle is moving.”

SEE: Tesla’s Autopilot: The smart person’s guide (TechRepublic)

If it works, Smith thinks this approach could “enable more advanced driving assistance or even higher levels of automation without the need for additional hardware.”

Radar has a long range, Smith said, which can help the vehicle operate at high speeds. Still, “relying solely on radar can produce a pretty distorted view of the world.”

Smith believes that “sensor fusion,” which takes data from multiple sources such as a camera, radar, and LiDAR could be a better way to ensure “perception actually matches reality.”

Miller agrees. “I think fusing the data together would be a better approach,” said Miller. “There have been many papers on this within the past few years since there is not one specific technology that dominates over the others. Incorporating radar and video together would produce a much better representation of the surrounding environment.”

Fleet learning

“The biggest development in this announcement is fleet learning,” said Smith. “Tesla is actually using its customers as trainers. These drivers will teach not only their own vehicles, but also all of the other Tesla vehicles on the road to correctly recognize roadway objects.”

SEE: Tesla’s Master Plan 2.0: AI experts, auto insiders, and Tesla customers weigh in (TechRepublic)

Smith said that other manufacturers and operators, like UPS and taxi fleets, should follow Tesla’s lead here. “Automated driving will progress much faster if the systems are learning from millions of ordinary drivers rather than a few hundred test drivers,” he said.

Preventing against driver abuse

In what seems to be a response to some documented reckless behavior by Autopilot users, Tesla will not let users re-engage Autosteer if they have ignored safety warnings.

Smith believes, however, that Musk should have emphasized the fact that Autopilot is an advanced driver assistance system–and should not be considered, in any way, a “self-driving car.” While Autopilot as a feature is impressive, it is more of an advanced cruise control that a car that drives itself. “Human eyes still remain a primary control sensor,” said Smith, “and I’m often concerned that drivers who casually learn about these developments might conclude otherwise.”

Tesla drivers respond

Daniel Nassarian, who uses Tesla Autopilot, doesn’t know how much it will impact the average Autopilot user. “I never really ran into any issues with the feature that would make me question its capabilities in the first place,” he said. “It seems like a seamless transition that most drivers won’t notice.”

Of course, it’s hard to say too much until he downloads the update, Nassarian said.

SEE: Tesla’s Musk says Autopilot update would have prevented fatal crash (ZDNet)

In thinking about Autopilot, it’s important to remember that the inherent purpose of the system is driver safety. Tesla is not the only company to install these features; automakers like Ford and GM have automatic braking and adaptive cruise control. And while Autopilot has been operating at the time of certain accidents, there are many cases in which it’s been shown to save lives–like when Joshua Neally suffered a pulmonary embolism and used Autopilot to help deliver him to the hospital.

The Autopilot update, Smith said, shows that Tesla is interested in communicating with the public. “I’m impressed that Tesla is being open,” said Smith, “about both its approach and the challenges inherent in that approach.”

Also see…