Wired
access:
Alibaba Cloud’s “Shadowless Cloud Computer” supports enterprises to quickly realize home office
access:
Apple Online Store (China)
Apple managers used drones to film the trip from Bozeman to the ski resort of Big Sky, so they could create a beautiful documentary featuring picturesque mountains as a backdrop CEO Tim Cook (Tim Cook) shows: Titan, the money-burning, multi-year project, is making progress.
Inside Apple, executives hailed the demo as a success: Tests showed the vehicles were driven without the high-definition 3D road map that most of its peers’ self-driving cars can’t do without them. Titan’s project managers hope to one day ditch these expensive and unscalable high-definition maps and build a truly self-driving car — one that can drive almost anywhere in the world without a steering wheel or pedals, the behavior of the car and the behavior of its passengers Security will be entirely the responsibility of Apple.
The good times don’t last long. Apple’s test car, a modified Lexus SUV, struggled to navigate the streets near its Silicon Valley headquarters without a map, sometimes hitting curbs and sometimes failing to drive properly, two people familiar with the project said. mouth. Earlier this year, a test car nearly hit a jogger crossing the street, a person familiar with the matter said.
The signs underscore the challenges Apple faces in terms of safety. Eight years after the project was established, the goals have been constantly changed, and the person in charge has also changed frequently. Behind the turmoil, software problems are much more difficult than expected, and the uncertainty of the listing date is one of the main reasons for the turnover of executives.
The demo also reflects a common business mistake that companies like Apple make when developing self-driving projects: wasting a lot of time on technology with minimal application scenarios, designing a demo along a specific route, using “what works here but not elsewhere”. The technology is almost useless, a phenomenon known as “demoware”. Some Titan developers say that Apple is much more involved in “demoware” than many of its peers.
Arun Venkatadri, who worked on self-driving cars at Uber, said: “As long as the money is available, the self-driving car can handle almost any fixed route. The question is whether the corresponding software can be scaled. Can it operate on more routes.”
To make matters worse, a project manager, Ian Goodfellow, recently left. He is a well-known machine learning expert who has been responsible for “infusing deep learning into Apple’s self-driving software.”
Apple’s plans to develop self-driving cars are almost an open secret in Silicon Valley. While many self-driving car developers have touted the “disruptive” technology, Apple has yet to publicly admit it. Google’s parent company Alphabet’s Waymo, General Motors’ Cruise, and Amazon’s Zoox want to operate Uber-like robotaxis (robotaxi) and roll out staggered peaks, while Apple’s goal is to sell cars. That means Apple’s challenge is far greater: Its self-driving doesn’t just work in specific areas, it has to work everywhere before anyone can buy it.
Apple’s self-driving test car based on a modified Lexus SUV
The companies have collectively spent more than $30 billion on related projects, based on public filings and accounts from some self-driving car developers, but have largely had no revenue so far. The technology is far from ready for large-scale operation, and no one seems to know how long it will take to be commercially available. It is equally unknown what breakthrough technology will be required to allow cars to predict the movements of cars, bicycles and pedestrians around them. It’s also a key challenge in getting self-driving cars to match or surpass human drivers.
But Tesla and some traditional automakers are already making billions of dollars a year from partial self-driving features like automatic cruise, lane keeping and automatic steering. This raises a question: Is Apple Car’s betting on fully autonomous driving a major business mistake?
“Cook Gene” makes the foundation of car building unstable?
Those bets mean a lot to Apple. Aside from its still-in-development mixed reality (MR) headset, the Titan is Apple’s biggest push into new territory since Jobs’ death in 2011. Project Titan is now led by Apple’s head of machine learning and AI, John Giannandrea, and has thousands of members.
The Titan and MR headset projects faced the same problem: Cook and Jobs had very different styles. Jobs dived into product design and pointed the way, while Cook and Titan kept their distance. Cook rarely visits Titan’s California office, according to multiple people involved in the project. Some former employees say that’s a big reason the project hasn’t taken off. Titan has never had a leader who can articulate the product, and frequent strategy changes have exhausted members. Cook’s reluctance to commit to mass production also discouraged some Titan executives.
Support for Titan from Apple’s executive team has also been mixed. Craig Federighi, senior vice president of software engineering, has been skeptical. He is well-known at Apple, responsible for software development, but not directly involved in the Titan project. Due to the frequent changes in leadership and strategy, and the ensuing layoffs, Titan became the object of ridicule from the rest of Apple’s business units. Several former employees said some managers had proactively warned employees to stay away from Titan.
Craig Federighi, Apple’s senior vice president of software engineering
Titan is very expensive, and the annual cost is more than 1 billion US dollars, but for Apple, it is a drop in the bucket. After all, the annual R&D expenditure exceeds 22 billion US dollars. In the eight years since the Titan project was established, Apple has spent more than $430 billion repurchasing shares. An Apple spokesman declined to comment.
Set sail towards the “physical vehicle”, and start to fight
In eight years, Titan has changed coaches four times, each with different ambitions. In 2014, when the Apple Watch was about to launch, Cook decided to branch out into electric vehicles. At the time, Tesla had been selling cars for six years, and Google had been testing self-driving cars for five years. Given Apple’s engineering and design prowess, building a car shouldn’t be difficult for Cook, two of the people said.
The Titan project has four core members, nicknamed “Four Kings” by Apple, namely Steve Zadesky, Benjamin Lyon, John Wright and DJ Novotney, who reported to Dan Riccio, then Apple’s head of hardware engineering. Among Apple’s executive team, Riccio is a major supporter of Titan.
Although Riccio is in charge of Titan, he doesn’t have much “intervention”, and a lot of work is in charge of Zadeski, who is the vice president in charge of Titan’s product design. Zadeski got to this position all the way from a mechanical engineer at Ford. He wanted to take Apple to build a traditional electric car and catch up with Tesla with more advanced “assisted driving” functions; but his colleagues have higher goals, such as fully autonomous driving. This is how infighting arises.
Apple has teamed up with luxury automaker Magna Steyr to build cars that were originally designed to resemble minivans. Apple was focusing on the passenger experience, not autonomous driving. Apple has envisioned how a car could detect a heart attack in a driver and send him to the doctor, as well as noise-cancelling surround sound.
In less than two years, Apple’s car-making has gone in many directions. Although the car-building project was only seen as a “technical investigation” at the time, Apple told employees that it hoped to sell cars as soon as 2019. Zadesky also hired auto industry experts to design doors, lighting and interiors, and began liaising with parts suppliers.
Half of Titan’s employees are from other teams in Apple’s consumer electronics division, and Apple’s move is a new way to build cars. Its industrial design team has put forward some radical ideas, such as whether it can be dominated by glass materials.
Apple wants to develop its own automotive engineering knowledge without relying on decades of experience in the industry, a decision that poses significant challenges. Engineers thought it was a waste of time.
Diverted to study software, the goal is “the iPhone of the automotive world”
At the end of 2015, after Cook’s shake-up in car building, facing new costs and safety risks, Titan stalled, and Zadeski chose to leave for personal reasons. About six months later, Riccio persuaded retired Apple executive Bob Mansfield to come back and reboot Titan.
Early Titan leaders Steve Zadesky (left) and Bob Mansfield (right)
After the change of leadership, Titan also suspended the development of the aforementioned cars and turned to fully autonomous driving software.
A big reason for this shift was the strong influence of Jonathan Ive, Apple’s head of industrial design at the time, who was adamantly opposed to making conventional electric cars. Ivey believes that autonomy is to Apple’s car, just as multi-touch is to the iPhone. If Apple wants to build a car, it has to create an iPhone in the mobile phone industry and overturn the entire industry.
It is worth mentioning that in mid-July, Ive and Apple’s 30-year partnership ended.
Former Apple chief designer Jonathan Ive (left) and senior vice president of hardware engineering Dan Riccio (right)
When Mansfield ended his partnership with Magna Steyr in 2016, Apple’s car had been redesigned into a compact BMW i3-like sedan. But self-driving software doesn’t yet know when it will be ready. Many Titan employees have been transferred to the “autonomous test vehicle” division (codenamed “Baja”) or the research and development of key vehicle components (battery, engine and powertrain).
bmw i3
In 2017, Mansfield believed that Apple was still a long way from perfecting its self-driving software, and decided to roll out the software in stages with relatively realistic goals, according to multiple people involved in the project. He focused on working with Volkswagen to develop a self-driving shuttle that would run the Infinite Loop from Palo Alto, Calif., to Apple headquarters.
That goal has since changed, however, to automating multiple bus routes that take employees to and from get off work around the Santa Clara Valley. Ultimately, Mansfield ended his partnership with Volkswagen, ditching the production of physical shuttles and focusing on test cars for the aforementioned routes.
The stumbling block is difficult to overcome, and the “software first” strategy is stranded
At this time, Apple has already fallen into “demoware”. The team invested in developing self-driving software, which was then demonstrated to executives. But two people involved in the development of the software said they were unable to expand or add functionality to the software.
Apple has made some progress in real-world testing. In 2017, Apple signed a lease-purchase agreement to take over a test track near Phoenix, Arizona, that once belonged to Chrysler. Apple bought the track last year through a shell company for $125 million, public filings show. In addition to simulations and testing on public roads, Apple is also testing self-driving software here. Apple built simulated streets on the track, using shipping containers as buildings, and hired contract workers to act as pedestrians or cyclists. This “three-pronged” approach is similar to that of friends such as Waymo.
Apple’s Robo City test track in Arizona
That same year, Apple’s first test cars began driving near its headquarters and on the Arizona race track. Modified Lexus vehicles cost upwards of $1 million each, largely because of the equipment they carry, including 14 Velodyne lidar sensors that use lasers to determine the distance of objects. (Apple later developed its own custom lidar, according to people familiar with the matter.)
But Mansfield’s leadership over Titan was limited. He was a former chip engineer with no experience in self-driving software or cars and could only give general directions, the people said. Mansfield is only returning to Apple on a part-time basis, with the ultimate goal of finding a full-time leader for Titan.
Retreat in spite of difficulties and re-study the physical vehicle
In 2018, Mansfield finally found the man: Doug Field. Field was Apple’s vice president of Mac hardware engineering and later spent five years at Tesla, where he co-designed the Model 3 and oversaw its production.
Like Zadesky, Field started his career at Ford as an auto veteran. Returning to Apple, Field laid off about 200 people from the Titan team and made designing physical cars a priority for the team again. Two people who worked with Field during this period said his goal was to sell cars, not run a robotaxi service. Field has automotive delivery experience but lacks software experience.
Apple makes cars, but the software is still immature.
In 2018, a former Titan engineer was sitting in a test car in Sunnyvale, Calif., following a simple route with intersections and T-junctions, when he discovered that the software was basically handing control to a backup human the entire time. Security Officer (This insurance is taken when the system doesn’t know what to do.)
Managers at Titan are more inclined to use deep learning as a way to solve self-driving software problems. Johny Srouji, Apple’s silicon chief, helped them build a custom chip, code-named “Tinos,” to train deep-learning models to help Apple’s self-driving software recognize objects and traffic signs. The process is called “perception”).
So, under the leadership of the third person in charge, Titan began to embrace deep learning.
Apple programmers hand-coded the rules for the car to follow, based on what the perception software detected. This approach is relatively risky and involves using deep learning to identify patterns in human driving style to determine the aforementioned rules and generate routes for cars. Experts express concern, saying the method remains unproven and may have to crunch huge amounts of data to come up with all possible rules.
In 2019, Apple hired Goodfellow, a widely respected deep learning researcher. He worked at Google, introducing deep learning to Titan’s “prediction and route planning” software. In 2020, Apple also assigned AI director Giannandrea, who also worked at Google, to oversee Doug Field and the Titan project, underscoring the emphasis on deep learning.
Apple deep learning researcher Ian Goodfellow
The executive presentation continues. Just before the outbreak of the new crown epidemic in 2020, Cook also took a test car in Silicon Valley and completed an uneventful trip. The car even underwent a DMV (California Department of Transportation)-certified drive test autonomously to demonstrate its capabilities, a person familiar with the matter said. Field told the team that the demo was a success.
Not every Titan member was shocked. In order to complete this demonstration, Apple engineers took a long time to thoroughly “sort out” this route with sensors such as lidar. But a person familiar with the matter said the car could not deviate from a preset route. If the self-driving software fails, a human driver will still have to take over.
Apple Car “landing” is nowhere in sight
Under the leadership of the third helm, Titan ushered in a period of stable development. Some former employees believe that Apple is likely to have successfully developed and released the car during Field’s tenure. But in September 2021, Field announced his resignation and returned to Ford as head of advanced technology and embedded software. Field, like Zadesky, struggled to get a commitment to mass production from Apple executives, two people familiar with the matter said. Although he can understand, Apple’s self-driving software doesn’t know when it will be ready.
Field’s resignation came after two senior Titan managers left, Lyon, who oversaw sensors, and Jaime Waydo, who oversaw operational security. At least three other Titan executives have left since Field’s resignation. In addition, some middle-level personnel left. That has raised questions about whether Apple can continue to develop cars.
Apple replaced Field with veteran Kevin Lynch, who previously oversaw software for the Apple Watch. In May, researcher Goodfellow told the team that he was resigning from Apple as the company asked employees to return to the office.
Kevin Lynch (left), the fourth head of Titan, and Doug Field, the third, (right)
The physical design of the Apple car continues. Designer Ive has told the Titan team that the focus should be on the vehicle’s unique design, not hidden sensors, a person familiar with the matter said. The current design features all four seats facing inward so passengers can talk to each other; and a curved roof similar to a Volkswagen Beetle.
Designers are also experimenting with a trunk. It can be raised to give owners more convenient access to the storage space in the trunk, and then lowered automatically. The designers also discussed the large seat screen, which rises from the back of the seat and lowers back into place when not in use. They also discussed the possibility of passengers sleeping on their backs in the car, although safety concerns may be waived. Apple also wants a “no steering wheel and brakes” waiver from the National Highway Traffic Safety Administration (NHTSA) and hired NHTSA chief legal counsel Jonathan Morrison to help last year, people familiar with the matter said.
Titan employees are discussing how to modify the new version of the self-driving car so that Apple can finalize the version and achieve the goal of hitting the road as early as next year.
“The Jogger Incident”
Apple is testing vehicles in cities with more rain and snow, including Los Angeles, San Diego, Seattle and around Lake Tahoe, collecting data on roads and local driving behavior. Earlier this year, Apple also planned to open operations in Portland, Las Vegas and Denver, a person familiar with the matter said. But it’s unclear when Apple will be ready to test its self-driving software in these new regions.
There are signs that Apple’s self-driving software is still immature. The removal of high-definition maps created a number of problems for test vehicles, according to data Apple filed with California regulators. In the test car, human safety officers frequently took over the car. In the first quarter of this year, an Apple test vehicle was traveling at 15 mph and nearly hit a jogger who was crossing the street on an unmarked crosswalk. Apple’s self-driving software recognized the jogger as a “stationary object,” then reclassified it as a “stationary person,” before successfully identifying it as a “moving pedestrian.” All this in less than a second.
But the car didn’t stop, just adjusted its course slightly. A person familiar with the situation said the car came to a stop just a few feet from a pedestrian thanks to a last-minute slam on the brakes by safety officers.
To this end, Apple suspended testing to investigate the so-called “jogger incident.” Testing resumed a few days later after the identification issues were resolved and the specific crosswalk was added to the map database.