Cruise’s head of artificial intelligence wants the autonomous-car startup to be defined by its AI innovation

Cruise AV
A Cruise AV in the Bay Area.

For autonomous vehicle startup Cruise, the future isn’t just about artificial intelligence. It’s about machine learning, and that’s why Cruise is teaching its electric vehicles to drive themselves in San Francisco – one of the most complicated urban environments for self-driving cars to operate in.

“Learning how to drive in San Francisco is amazing for AI,” said Hussein Mehanna, the company’s head of AI, noting that the dense and unpredictable streets are ultimately an advantage. “The more interesting the data, the more the machine can learn.”

Mehanna hopes that learning will not only revolutionize autonomous driving, but also plant Cruise at the forefront of the next big thing: AI-based companies.

Taking machine learning to a new level

General Motors bought Cruise back in 2016 for around $1 billion, and through subsequent investment rounds, it’s grown to a nearly $30 billion valuation. The company’s goals are spectacularly ambitious, with CEO Dan Ammann effectively calling for the end of personal-car ownership and spurring Cruise to go after a multi-trillion-dollar future global ride-hailing opportunity.

In order to get there, Cruise needs game-changing hardware and software – a quest overseen by Kyle Vogt, its cofounder and chief technology officer – and high-profile partners, including ones it already has like GM and Honda. But Cruise also needs artificial intelligence and machine learning at a level that, frankly, nobody has seen before.

Hussein Mehanna Headshot
Hussein Mehanna.

As powerful as 21st-century AI sounds, Mehanna said it’s only recently that its full capabilities have been unleashed. Advancements in robotics and machine learning have made that possible.

“I always had a fascination with AI,” Mehanna, whose career path to Cruise included stints at Facebook and Google, told Insider in an interview. But where are all the robots we might have expected to see by now?

Mehanna said the kind of AI we see in demonstrations – dancing humanoids robots on YouTube, for example – doesn’t scale.

“They’re scripted to handle a certain number of use cases,” he said.

cruise self driving car san francisco people walking
A Cruise vehicle in San Francisco in May 2019.

Enter machine learning, which he said has the critical power to generalize.

This is, to put it mildly, huge. At Cruise, Mehnna’s team is tackling a whole new way of undertaking computer science, led by those autonomous EVs cruising through San Francisco.

If it all comes together and Cruise is able to successfully commercialize its service, then Mehanna said that the company could notch an unprecedented achievement: becoming what he termed the first “AI-native company.”

Dreaming of robots that can do much, much more

“It’s a new concept, and we’re inventing it,” he said. The analogy that leaped to mind for him was being able to handle HTML coding for the internet of the late 1990s.

“If you knew HTML, you were a rocket scientist,” he said. The skillset led to internet-native companies such as Google. That history is now staged to repeat with Cruise.

“In five to 10 years, AI natives will be the status quo,” he said.

The endgame of this process should be what he called a “general-purpose robot,” able to learn as humans now learn. It could drive a car, fly a plane, or attend to more mundane tasks.

“My dream,” he said, “is to get my laundry folded by a robot.”

Walmart Cruise self driving car

Talking to Mehanna, one gets that sense that we’re just at the beginning of something radical in changing how the world operates. Cruise has already made huge leaps in teaching a car to drive itself, once the stuff of science-fiction movies. But for Mehanna, those apparent leaps are but small steps toward robotic applications and machine learning remaking numerous aspects of everyday life – aspects that we take for granted or have long assumed would always have to involve natural, rather than artificial intelligence.

In the short term, however, he’s simply contemplating machine learning as a prerequisite to Cruise accomplishing what it set out to do five years ago.

“At Cruise, you can’t have a company without AI,” he said.

Read the original article on Business Insider

As Tesla gears up to release ‘Full Self-Driving’ to thousands more drivers, videos continue to show its flaws

Tesla Model S interior
Tesla Model S interior.

  • Tesla is eyeing next month for a wider release of its “Full Self-Driving” beta software.
  • The system is still buggy and sometimes gets drivers into dangerous situations, videos show.
  • Neither the software nor Tesla’s Autopilot make its cars fully autonomous, despite the company’s marketing.
  • See more stories on Insider’s business page.

When Tesla beamed out a prototype version of its “Full Self-Driving” (FSD) technology to select Tesla owners in October, videos of the driver-assistance system fumbling normal traffic situations – from failing to follow road rules to nearly steering into parked cars and cement barricades – flooded the web.

Now Elon Musk wants any Tesla owner that has paid for FSD to have access to the beta next month. But clips cropping up online continue to cast doubt on whether the technology is safe enough to test on public streets.

A March 12 video posted by Youtube user AI Addict shows a Model 3 running FSD beta version 8.2 clumsily navigating around downtown San Jose at dusk. With FSD switched on, the vehicle nearly crashes into a median, attempts to drive down railroad tracks, and almost plows down a row of pylons separating the road from a bike lane. All of those dicey situations were narrowly avoided only because the driver quickly took over control.

In another clip posted March 18, Model Y owner Chuck Cook tests the beta’s ability to make unprotected left turns. The software performs admirably a few times, waiting until a break in traffic to cross the three-lane road. More than once, however, Cook has to slam on the brakes to avoid coasting into oncoming traffic. And on his last go, the Tesla nearly drives headlong into a pickup truck with a boat in tow.

FSD testing videos have become an entire genre on YouTube. Many of them depict cars comfortably navigating lane changes, four-way stops, and busy intersections. Yet the buggy clips illustrate the potential dangers of letting amateur drivers experiment with a prototype software on public roads.

Tesla is using its owners as “guinea pigs for the technology,” Jason Levine, executive director of the Center for Auto Safety, a consumer advocacy group, told Insider. “And what’s much more concerning, quite frankly, is they’re using consumers, bystanders, other passengers, pedestrians, and bicyclists as lab rats for an experiment for which none of these people signed up.”

FSD – a $10,000 add-on option – is a more advanced version of Tesla’s Autopilot, its standard driver-assistance feature that enables cars to maintain their lane and keep up with highway traffic using a system of cameras and sensors. FSD currently augments Autopilot with features like self-parking, traffic light and stop sign recognition, and the ability to take highway on-ramps and exits.

The limited beta software in question adds on a capability critical for any system that aims to be called fully self driving: the ability to navigate local streets, which, as opposed to highways, have a much more complex driving environment that includes left-hand turns across traffic, pedestrians, cyclists, and the like.

Even before it introduced the FSD beta last fall, Tesla faced scrutiny over Autopilot and its potential for abuse. The National Highway Traffic Safety Administration confirmed earlier this month that it is investigating Autopilot’s role in 23 recent crashes, including multiple where Teslas barreled into stopped emergency vehicles. Over the years, numerous videos have surfaced on social media of drivers sleeping with Autopilot turned on or otherwise misusing the feature.

Read more: Tesla and Apple are incredibly important companies, but their progress on self-driving vehicles is pathetic

To make things safer, Levine said, Tesla could begin using vehicles’ internal cameras to monitor driver attention as many other carmakers do. Currently, Tesla only monitors whether a driver’s hand is on the steering wheel, while other systems, like GM’s Super Cruise, track a driver’s eyes to make sure they are paying attention to the road.

Changing the names of Autopilot and FSD – which are misleading since neither technology is autonomous and both require constant driver attention – would be a good start as well, Levine said.

“The insistence on utterly hyperbolic description really undermines any sort of good-faith effort to present this technology in a way that is going to not present an unreasonable risk,” he said.

For Tracy Pearl, a law professor at the University of Oklahoma who researches self-driving technology, the main problem isn’t so much the quality of Tesla’s driver-assistance systems, but rather the way drivers interact with them.

Although advanced driver-assistance suites like Tesla’s could make cars safer when used properly, research has shown that drivers on the whole don’t understand their capabilities and limitations, Pearl said. Moreover, drivers’ attention tends to wander when those features are switched on. Tesla exacerbates these issues by marketing its tech in ways that overstate the cars’ abilities, but the information gap between manufacturers and drivers extends to other carmakers as well, she said.

Problems with the way driver-assistance systems are marketed and the way drivers interact with them are heightened where a beta system is concerned, she said.

“I think calling it Full Self Driving is not just deceptive, I think it is an invitation for people to misuse their cars,” she said.

Tesla, though it consistently claims that self-driving cars are right around the corner, acknowledges that both Autopilot and FSD require constant driver attention and aren’t fully autonomous. Still, it’s much more blunt with regulators than it is with the general public.

In a series of emails to California’s Department of Motor Vehicles in late 2020, a Tesla lawyer said that FSD can’t handle certain driving scenarios and that it will never make cars drive themselves without any human input.

Tesla also makes an effort to inform users about the risks of using a system that isn’t completely ready for prime time. In a disclaimer beta testers received with the October update, Tesla urged drivers to use the software “only if you will pay constant attention to the road, and be prepared to act immediately.

“It may do the wrong thing at the worst time,” Tesla said.

Tesla did not respond to a request for comment for this story.

Read the original article on Business Insider

Tesla says ‘full self-driving’ beta software will never be fully autonomous, emails to regulators show

Tesla Model 3_2
Tesla’s FSD beta will remain an advanced driver-assistance system, not a fully autonomous one.

  • Tesla told regulators its “full self-driving” beta won’t be fully autonomous when it’s released.
  • Emails between Tesla and California’s DMV about the software were made public on Friday.
  • The emails appear to contradict some of what CEO Elon Musk has said about the tech.
  • Visit the Business section of Insider for more stories.

From launching a fleet of robotaxis to navigating a cross-country road trip without human intervention, Tesla boss Elon Musk has been touting ambitious plans to roll out self-driving cars for years.

And although Tesla has been pre-selling a $10,000 feature called “full self-driving capability” (FSD), the carmaker has told California regulators it won’t make cars fully autonomous in the immediate future, according to emails made public on Friday.

In a December 14 email to California’s Department of Motor Vehicles, Tesla lawyer Eric Williams said the company “made it abundantly clear” to beta participants that the system “does not make the vehicle autonomous and that the driver is responsible for being fully attentive at all times.”

Tesla released a key element of FSD known as “autosteer on city streets” to a limited number of company employees and owners as part of a beta program in October, with plans to widely release the system in the early part of 2021.

Williams went on to say that “we do not expect significant enhancements” that would allow the system to take on all driving tasks, and that “a final release of City Streets will continue to be an SAE Level 2, advanced driver-assistance feature.” Level 2 autonomy means that a vehicle needs constant human attention.

The emails were obtained and published by the transparency advocacy group PlainSite, which has routinely sparred with Tesla in the past. The DMV reached out to Tesla about the FSD beta’s capabilities to make sure the carmaker did not plan to test an autonomous vehicle system on public roads without the proper permit.

In response, Williams argued in a November 20 email that the beta has significant limitations that disqualify it from being regulated like an autonomous system.

“The driver must fully supervise the system,” Williams said, as it is not meant to navigate “certain circumstances and events” such as “road debris, emergency vehicles, construction zones, large uncontrolled intersections with multiple incoming ways, occlusions, [and] adverse weather.” Those are all situations that a more advanced Level 3 system would be expected to navigate or recognize.

Tesla’s official stance that the FSD beta’s public release won’t provide full autonomy appears to contradict Musk’s bullish claims about the technology.

In December, Musk tweeted about Tesla’s system enabling one to drive while playing video games. On a conference call in January, Musk indicated Tesla would achieve Level 5 autonomy, which requires zero human intervention, in 2021. Tesla’s CEO also said that FSD would enable the company to launch a fleet of 1 million robotaxis by 2020, which did not materialize.

However, “autosteer on city streets” is just one feature of the FSD package, so it’s possible the system will expand or evolve at some point in the future to take on more driving tasks.

The release of FSD has been delayed numerous times, but Musk said in March that it will be available as a subscription before July.

Tesla did not respond to a request for comment.

Read the original article on Business Insider

Toyota just started building a 175-acre smart city at the base of Mount Fuji in Japan. Photos offer a glimpse of what the ‘Woven City’ will look like.

Toyota city
The “Woven City” will eventually be home to 2,000 Toyota employees and their families, retired couples, retailers, and scientists, according to the company.

Toyota Motor Corporation started construction this week on a 175-acre smart city at the base of Japan’s Mount Fuji, about 62 miles from Tokyo, the company announced Tuesday.

The city, which Toyota has dubbed the “Woven City,” is expected to function as a testing ground for technologies like robotics, smart homes, and artificial intelligence. A starting population of about 360 inventors, senior citizens, and families with young children will test and develop these technologies.

These residents, who are expected to move into the Woven City within five years, will live in smart homes with in-home robotics systems to assist with daily living and sensor-based artificial intelligence to monitor health and take care of other basic needs, according to the company.

The eventual plan is for the city to house a population of more than 2,000 Toyota employees and their families, retired couples, retailers, and scientists. Toyota announced plans for the city last year at CES, the tech trade show in Las Vegas.

Here’s what the 175-acre smart city is set to look like when it’s finished.

Toyota’s planned 175-acre smart city will sit at the base of Mount Fuji in Japan, about 62 miles from Tokyo.

toyota city
An artist’s rendering of Toyota’s planned smart city.

Called the “Woven City,” the development will feature pedestrian streets “interwoven” with streets dedicated to self-driving cars, according to press materials. The city is expected to be fully sustainable, powered by hydrogen fuel cells. 

The Woven City will function as a testing ground for technologies like robotics, smart homes, and artificial intelligence, according to the company.

Toyota officially started construction on the city in a groundbreaking ceremony on Tuesday, the company announced. The city is set to be built on the site of one of Toyota’s former manufacturing plants called Higashi-Fuji.

Toyota plans to send about 360 people to live in the Woven City to start. From there, it intends to gradually grow the population to more than 2,000.

toyota city
An artist’s rendering of Toyota’s planned smart city.

The first residents will be a group of roughly 360 inventors, senior citizens, and young families with children, according to the company. These residents will move in within five years, a Toyota spokesperson told Insider last year.

Toyota has not yet revealed how these first residents will be chosen, and a spokesperson did not immediately respond to Insider’s request for more details.

Eventually, the Woven City is expected to be home to more than 2,000 Toyota employees and their families, retired couples, retailers, visiting scientists, and industry partners.

Residents will live in homes outfitted with in-home robotics technology as well as sensor-based artificial intelligence to monitor their health and take care of their basic needs.

Toyota city
An artist’s rendering of a home in Toyota’s planned smart city.

Despite the planned high-tech homes, Toyota says that promoting human connection is a major theme of the city but has not released specifics on how it plans to encourage this. 

Press materials indicate that the planned city will feature multiple parks and a large central plaza for social gatherings.

toyota city
An artist’s rendering of Toyota’s planned smart city.

Bjarke Ingels, the famed Danish architect behind high-profile projects such as 2 World Trade Center in New York City and Google’s California and London headquarters, is responsible for the city’s design.

Buildings are to be made mostly of wood to minimize the carbon footprint.

toyota city
An artist’s rendering of Toyota’s planned smart city.

Rooftops are slated to be covered in photo-voltaic panels to generate solar power and hydrogen fuel cell power.

Toyota says it plans to integrate nature throughout the city with native vegetation and hydroponics, a method of growing plants without soil.

The city will be designed with three different types of streets: one for self-driving vehicles, one for pedestrians using personal mobility devices like bikes, and one for pedestrians only.

Toyota city
An artist’s rendering of Toyota’s planned smart city.

These three types of streets will form an “organic grid pattern” to help test autonomy, according to Toyota.

There will also be one underground road used for transporting goods. 

A fleet of Toyota’s self-driving electric vehicles, called e-Palettes, will be used for transportation, deliveries, and mobile retail throughout the city.

toyota e-palette

Toyota has not yet disclosed an estimated completion date or estimated total cost for building the Woven City. 

The Woven City joins a slew of similar smart city projects across Japan, some of which are also spearheaded by major companies.

panasonic Fujisawa Sustainable Smart Town
A robot demonstrates a delivery at Panasonic’s Fujisawa Sustainable Smart Town in Fujisawa, Japan on December 9, 2020.

In 2014, electronic appliance company Panasonic opened a smart city in Japan’s Kanagawa Prefecture called the Fujisawa Sustainable Smart Town, per Tokyo Esque, a market research agency. The city is still under construction with completion expected in 2022, but more than 2,000 people live there now, according to Panasonic.

Accenture, an American-Irish consulting company, is teaming up with the University of Aizu on smart city projects in the town of Aizuwakamatsu with the goal of better using artificial intelligence in public services, the company announced in July 2020.

Local governments made more than 50 proposals for smart cities in Japan in 2020, but only a handful of those were approved, according to Tokyo Esque.

As Linda Poon reported for Bloomberg last year, critics say smart city developers should focus on the human aspect of the projects, not just the technology.

“If it’s not started from a human-centric perspective, from the bottom up as opposed to from the top down, these aren’t real cities,” John Jung, founder of the Intelligent Community Forum think tank, told Bloomberg in January 2020. “They’re not designed to get [people] to know each other.”

Read the original article on Business Insider

One of Uber’s earliest investors says the billions it spent on self-driving were a ‘waste of money’

Uber ATG Self driving   Photo by
Uber launched its own self-driving team in 2015.

  • Early Uber investor Bill Gurley said the company “burned” billions of dollars on self-driving tech. 
  • That money would have been better spend on food delivery, he told journalist Eric Newcomer. 
  • Uber’s sold off its self-driving unit in late 2020 as it grappled with a revenue downturn.  
  • Visit Business Insider’s homepage for more stories.

Uber’s food-delivery business has been a massive boon for the company as the pandemic sent ride-hailing revenues down the tubes and takeout orders through the roof.

One early investor in the company, Benchmark’s Bill Gurley, says the company should have committed more investment to food delivery that it instead spent trying to build self-driving vehicles. 

In an interview with journalist Eric Newcomer in his newsletter Friday, Gurley lamented that Uber dumped so much capital into its self-driving unit, called the Advanced Technologies Group (ATG). Gurley’s firm, Benchmark Capital, first invested in Uber in 2011, according to Pitchbook data. Gurley sat on Uber’s board until 2017. 

“We probably burned $2.5 billion on autonomous that was a waste of money,” Gurley said, adding that in retrospect that sum would have been better spent on growing Uber Eats. 

However, he said, there still would have been major risks to investing heavily in delivery. 

“These ideas where you use capital as a weapon to build liquidity and then emerge out of it later is a really hard and dangerous game, and you need the capital markets to support it,” Gurley said. “And if we hadn’t gone from a frothy market five years ago to a super frothy market now, maybe you don’t make that gap.”

Uber has historically struggled to turn a profit – it reported an adjusted net loss of $8.5 billion in 2019, for example – leading it to double down on core businesses and trim some capital-intensive side projects as the pandemic slashed revenues. In December, the company announced plans to sell its ATG unit to the competing autonomous-vehicle firm Aurora, and its flying taxi program to Joby Aviation.

Read more: Uber ATG has been hobbled by a deadly crash, infighting, and balky tech – and investors are losing patience with the self-driving division

The appeal of self-driving vehicles for a ride-hailing service is clear: if Uber could develop robotaxis capable of reliably ferrying passengers around without intervention, it would theoretically save heaps on payroll and other costs associated with employing human drivers. But that’s much easier said than done. 

Since its founding in 2015, Uber ATG struggled to make significant headway toward building an autonomous taxi, falling behind competitors like Alphabet’s Waymo and GM’s Cruise. The firm was plagued by infighting, allegations that its lead engineer stole trade secrets from Google (he was later convicted and pardoned by President Donald Trump), and a fatal crash in 2018 that’s considered to be the first deadly incident involving a self-driving car. 

Read the original article on Business Insider