It is no secret that Tesla's autopilot project is fighting. Last summer, we reported a report that Tesla bleeding talent from his autopilot department. Tesla autopilot chief Sterling Anderson has left Tesla at the end of 2016. His successor was Chris Lattner, who had previously developed the programming language Swift at Apple. But Lattner only lasted six months before he left last June.
Now leaves Lattner's successor Jim Keller Tesla.
Keller was a well-known chip designer at AMD before being hired for Tesla's hardware development for autopilot in 201
Kellers departure comes just weeks after the death of Walter Huang, a Model X driver in Mountain View who crashed into a concrete scout, California. Tesla said autopilot was busy at the time of the crash. Since then, Tesla has fallen into public feuds with both Huang's family and the National Transportation Safety Board, the federal agency investigating the crash.
"Today Jim is Kellers last day at Tesla, where he has low-voltage hardware, autopilot software and infotainment," Tesla said in a statement to Electrek. "Before joining Tesla, Jim's core passion was the microprocessor generation, and he joins a company where he can focus exclusively on it."
Tesla has a major autopilot problem
We do not know why Keller has decided to leave. What we do know, however, is that he was facing an extremely challenging situation in Tesla.
Over the last 18 months, Tesla has charged its customers $ 3,000 for a "full self-driving" option for autopilot. Tesla has promised customers that their current hardware – which includes radar and a range of cameras but no lidar – will be enough to achieve full self-drive capability. And since June 2016, Elon Musk predicts that full self-drive capabilities will be available in the next two years.
But almost two years after Musk's prediction, Tesla does not seem to be close to self-drive skills for customers who have paid for the feature. And it's a real question if it's even possible to deliver full self-propelled capabilities with the hardware that Tesla has delivered since 2016.
Any other large-scale, self-driving technology company uses Lidar as a key element of its sensor suite. Musk has repeatedly dismissed Lidar as a "crutch," but it seems that companies like Waymo and GM know what they're doing. Tesla's cars have less hardware redundancy than the Waymo and GM cars, leading to potential safety concerns when key components fail. And it's not known if the Nvidia processors on current Tesla cars are even powerful enough to drive themselves.
In fact, the Wall Street Journal Sterling Anderson, the head of the autopilot program, when Tesla began to sell the "full self-propelled" property, was not sure that the hardware would be sufficient for full self-propelled skills. "The announcement shook some engineers because they believed the product being released was not self-propelled," Journal said. When Anderson was asked about moving an employee, he reportedly replied, "This was Elon's decision."
The decision to market autopilot as a "fully self-propelled system" was a factor in Anderson's decision and at least two other engineers to leave the company, " Journal reported.
Since The project has suffered a steady loss of talent in its August 2008 article Journal noted that the team has "lost at least 10 engineers and four top executives." That included Anderon's successor, Chris Lattner, who only lasted six months.
And Lattner was not subtle when he left Tesla last summer.
Turns out Tesla does not suit me well, I'm interested in interesting roles for an experienced technical Ladder!
– Chris Lattner (@clattner_llvm) June 21, 2017
Why the autopilot team is so tough
According to Electrek, Tesla plans to divide Keller's role into two jobs. Chip designer Pete Bannon succeeds Keller as head of autopilot hardware efforts. In the meantime, Andrej Karpathy, head of Autopilot's machine vision efforts, will be promoted to head all autopilot software efforts.
Keller's successors will be subjected to immense – and possibly contradictory – pressure.
On the one hand, the autopilot team is facing pressure from regulators and safety representatives to position the autopilot more clearly as a driver assistance feature. This could mean that the autopilot function is further restricted without the driver having his hands on the steering wheel – or reducing the number of warnings a driver receives before the technology shuts off. It could mean that new techniques are added to make sure the driver is paying attention. For example, a camera facing the driver. And it could mean restricting the autopilot's operations on roads where Tesla is sure it's safe.
At the same time, the autopilot team is under pressure from Musk – and customers who paid $ 3,000 for full self-drive Expand the capabilities of the autopilot and finally enable hands-free operation on all roads. Ideally, they would find out how to do this without the need for LIDAR sensors or other expensive upgrades.
The problem is of course that this is not possible.
In fact, Google – now Waymo – concluded years ago that the evolutionary approach that Tesla is currently pursuing makes no sense. Engineers found that test drivers relied too soon on early versions of their self-propelled technology, even though they were explicitly told to keep their hands on the wheel and their eyes on the road. So Google rejected plans to market a driver assistance technology and instead focused on building completely self-driving cars where a customer would never have to touch the steering wheel.
Waymo has focused on building an autonomous taxi service because it allows the company to roll out service step by step, city by city. Waymo plans to launch in Phoenix, where 99 percent of the time there are pristine roads and perfect weather. Over time, as the software becomes more powerful, service is extended to areas of more difficult situations, such as snow or high-density urban environments. This strategy means Waymo can add or replace sensors and other hardware to its cars while learning what is needed for full self-drive.
But this strategy does not go well with a company like Tesla, which sells cars directly to customers. Customers expect their cars to work everywhere, not just in a specific metropolitan area or only in favorable weather conditions. Tesla does not have billions of dollars to map the city streets and extensively test the software with security drivers. And customers who have already paid for the full self-drive mode will no longer want to pay thousands of dollars to add Lidar or additional computing power.
In short, Bannon and Karpathy have a challenging task ahead of them. If Elon Musk does not want to lose the autopilot leader all the time, he might consider changing Tesla's autopilot strategy to make sure the job is not impossible.