Tesla needs to rethink its Autopilot technology

15.07.2016
In light of several accidents that have involved its assisted driving technology, Tesla should rethink just how much automation it provides to drivers, automotive experts said today.

While Tesla's Autopilot advanced driver assistance system (ADAS) uses cameras and radar, the carmaker should consider adding LiDAR (Light Detection and Ranging) technology, which uses lasers to create 3D scans of objects around a vehicle.

Currently, Tesla's radar only ranges the distance to objects, but it cannot necessarily determine that those objects are.

Tesla CEO Elon Musk has said he doesn't think his all-electric vehicles need LiDAR, and argues that passive forward-looking radar can accomplish ADAS functions.

"I think that completely solves it without the use of LIDAR. I'm not a big fan of LIDAR, I don't think it makes sense in this context," Musk said last year.

More recently, there has been speculation that Tesla is at least considering adding LiDAR at some point.

But, more may be needed.

Richard Wallace, the director of the Transportation Systems Analysis group at the Center for Automotive Research, said if Tesla vehicles were equipped with vehicle-to-vehicle (V2V) and vehicle-to-infrastructure (V2I) communications technology, the recent spat of accidents would likely have been avoided.

Wallace believes that autonomous vehicles won't be entirely safe until they can  communicate with other vehicles on the roads as well as the infrastructure around them.

"People argue...'I developed a fully self-driving vehicle based purely on sensors and artificial intelligence and I don't need V2V and V2I technology,' " Wallace said. "V2V would have stopped this...fatal crash."

For the past two weeks, Tesla has been at the heart of a media feeding frenzy since it revealed that an owner of one of its Model S sedans was killed back in May while the Autopilot feature was engaged. That accident was followed by two others allegedly involving Autopilot.

The National Highway Traffic Safety Administration (NHTSA) and the Securities and Exchange Commission are investigating Tesla over the incidents.

Last week, the NHTSA requested additional data from Tesla regarding Autopilot, including any design changes and updates to the system, as well as detailed logs of when the system has prompted drivers to take over steering.

Today, Consumer Reports, whose past reviews showered Tesla's all-electric vehicles with praise, called on the all-electric carmaker to disable its semi-autonomous driving system in light of the accidents.

"By marketing their feature as 'Autopilot,' Tesla gives consumers a false sense of security," Laura MacCleery, vice president of consumer policy and mobilization for Consumer Reports, said in a blog. "Consumers should never be guinea pigs for vehicle safety 'beta' programs."

Wallace agreed that Autopilot misrepresents what the technology offers.

"It's driver assistance at this point," Wallace said.

Autopilot has been described by Tesla as a public beta program that's intended to assist and not fully take over the task of driving. Data from the beta program is transmitted back to Tesla, allowing the company to improve the technology.

Musk said Tesla had no plans to disable Autopilot, ardently defended it, and reiterated that the company plans to blog about how to properly and safely use it.

"This is the first known fatality in just over 130 million miles where Autopilot was activated. Among all vehicles in the U.S., there is a fatality every 94 million miles. Worldwide, there is a fatality approximately every 60 million miles," Musk said in a blog.

John Dolan, principal systems scientist at Carnegie Mellon University's (CMU) Robotics Institute, said it's a bit "glib" to compare miles driven by humans and miles driven while an ADAS is engaged; it's not an apples-to-apples comparison.

Dolan pointed out that when Autopilot is engaged, there may be many instances when the driver is forced to take control of the vehicle to avoid an accident. To insinuate that Autopilot is safer isn't necessarily accurate.

"I haven't seen exactly what the nature of the data is. If you've only logged so many miles but not the human interventions...it's hard to say this validates it," Dolan said. "What I'd like to see is more information about the information."

Tesla has been adamant that Autopilot is an advanced driving assistance system (ADAS), and not fully autonomous driving technology, meaning drivers must keep their hands on the steering wheel and be alert at all times. The Internet, however, is rife with Telsa vehicle owners who recorded their decision to ignore that advice by driving without their hands on the wheel, and in several cases have even climbed into the back seat of their vehicles to prove their confidence in the technology.

Currently, Tesla's Autopilot system conforms to level 2 (or slightly higher) autonomous functionality, according to the NHTSA. The agency has created five levels to describe autonomous functionality, level 0 equating to no self-driving features and level 4 being a fully self-driving vehicle.

Autopilot does go beyond ADAS technology offered by other carmakers in that it can automatically change lanes when the turn signal is activated.

Tesla's vehicles use both radar and cameras to feed data to a central computer, which uses algorithms to enable Autopilot to manipulate the vehicle for functions such as adaptive cruise control, lane centering or automatic braking.

Both Google and Tesla have "lit a fire under" the auto industry to forge ahead more aggressively with ADAS technology, Dolan said. Virtually all carmakers today have or are planning to include ADAS on most vehicle models.

CMU has been developing autonomous vehicles for years, pushing out more than a dozen models; the most recent uses LiDAR.

LiDAR systems, however, paint a more robust picture that can be used by a vehicle's artificial computer algorithms to better determine its surroundings. One problem with rolling out LiDAR, however, is its exorbitant price.

LiDAR systems, such as the ones being tested by Google on its autonomous vehicles, can cost more than $1,500. As the technology evolves, prices are expected to come down quickly. Some LiDAR makers are already designing systems that will cost as little as $250.

A better question may be: "Are we ready for autonomous driving technology" Dolan said.

The problem is that drivers get lulled into a false sense of security when they believe a computer can perform better than they can.

On one hand, Dolan applauds Musk and Tesla's bold approach to deploying self-driving technology because it is driving development. On the other hand, ADAS that's not fully baked and ends up directly or indirectly causing accidents, could encourage  the industry as a whole to pull back hard on the deployment throttle.

While the rest of the auto industry may be developing its own ADAS and fully autonomous vehicle technology, it's not likely to take the same path as Telsa. The 100-plus-year old auto industry is far more cautious because, unlike a tech start-up, it has more to lose, Dolan said.

"The last thing they want is the publicity that Tesla is getting right now, and they'll do anything to avoid it," he said.

(www.computerworld.com)

Lucas Mearian

Zur Startseite