Innovation Artificial Intelligence

Tesla's Master Plan 2.0: AI experts, auto insiders, and Tesla customers weigh in

Tesla has just unveiled its second 'master plan,' coming a decade after the first. Here's what the autonomous car insiders think about what it means and whether it's on target.

Image: Hope Reese/TechRepublic

A decade after Elon Musk announced Tesla's first master plan, he has just come up with its sequel--a grand outline that has been called out for being "absurd," and "insane," while simultaneously "brilliant" and "magnificent."

So, which is it?

The press and public often seem to have a difficult time synthesizing Musk's lofty ambitions. Is Tesla even an automaker? Or just a tech company? The announcement of the new master plan arrived days after Tesla officially changed its website from teslamotors.com to tesla.com--which could signal a conceptual shift to the digital world.

While many laud Musk for his efforts in getting electric cars on the grid--which will switch into high gear with Tesla's release of the $35,000 Model 3 next year--he has also been criticized for getting in over his head when it comes to moving towards an autonomous driving ecosystem. This criticism that came into full focus with the recent fatality that occurred in a Tesla Model S operating in Autopilot mode.

What's included in Tesla's new master plan? Here are the highlights:

Bring together car and battery. In order to "create stunning solar roofs with seamlessly integrated battery storage," Tesla plans to integrate with SolarCity. The merger is already in the works, which will eventually enable Tesla's Powerwall batteries to be powered by SolarCity's solar panels.

Autonomous driving. Tesla is currently offering Autopilot, an advanced driver assist mode that can self-steer, brake, and switch lanes. But once the technology is ready, it will "develop a self-driving capability that is 10X safer than manual [driving] via massive fleet learning."

SEE: Tesla driver dies in first fatality with Autopilot: What it means for the future of driverless cars (TechRepublic)

Buses and trucks.Tesla plans to tackle "the major forms of terrestrial transport," meaning busses and trucks.

The fleet. Tesla envisions a shared system of car ownership in which it operates a "fleet" of autonomously driving cars that can be summoned to pick up passengers, reducing cost and transforming our traditional concept of ownership

So what do the experts think? To suss out the facts from the fantasy, TechRepublic spoke to several insiders in the driverless car world.

The big picture

Bryant Walker Smith, professor at the University of South Carolina School of Law and School of Engineering, is one of the leading thinkers in the autonomous vehicle world with insight into the legal aspects of driverless vehicles on the road.

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

Smith has "no criticisms" about Tesla Master Plan part deux. "It's good that Tesla is sharing its vision with the public," he said. "In other contexts, Tesla does need to be more concrete and share more of the data and analyses that underlie its conclusions. But this is not one of those contexts."

Smith also wonders if Tesla will "explore micro-trucks, delivery robots, and other forms of more localized (and necessarily automated) shipping," not to mention drones. He also said he "wonder[s] how Tesla will play in the digital world. For example, will it monetize the data it is collecting about its vehicles, their users, and the environments in which they operate?" Smith asked.

He also sees Tesla's approach as "relatively conservative," in terms of the legal complexity of the "more advanced forms of driving automation," which surprises him.

Overall, however, Smith sees Tesla as doing the right thing here. "Developers of automated systems should be publicly sharing and substantiating their safety philosophies," Smith said.

Adoption of the fleet concept

Evan Fusco, administrator for Tesla's online forum, "had no idea that the mass transport and semi-truck elements were on the drawing board."

"I hate to see projects like that get in the way of implementation of the Model 3 mass market rollout, but perhaps it won't," Fusco said.

In terms of the autonomous fleet, Fusco still sees many technological challenges and wonders how ordinary drivers will respond.

"I don't think I want strangers riding around in my car, paid or otherwise, while I'm off on vacation," said Fusco. "But perhaps there's a segment who want a car and can make the case that by having it as a loaner most of the time that they essentially will get the car for free while others pay for it."

Driving data

Musk noted that we would need six billion miles of autonomous driving data in order for "worldwide regulatory approval." At the rate we're at now, we get three million miles a day.

Jim Adler, head of data at Toyota Research Institute said that "it's far too early to estimate what regulatory parameters will be deemed appropriate. But every year, 50 million new cars are sold. At 10,000 miles per car, these new cars drive 500B miles in their first year. That's 83 times 6 billion."

SEE: How driverless cars will transform auto insurance and shift burden onto AI and software (TechRepublic)

Still, Adler said these kinds of miles "are probably not the "right" kind of miles. They need to include very demanding and hazardous situations, all weather conditions, regional driving styles, etc," said Adler. "That's why simulation will be so important to put every software change through the "right" kinds of driving environments."

Tech challenges

John Dolan, a principal systems scientist in the Robotics Institute at Carnegie Mellon University, sees some tech challenges with the plan. Here are three:

  1. Achieving good enough performance from sensors that are sufficiently inexpensive. Dolan said the first challenge is the fact that GPS systems are currently priced in the tens of thousands of dollars, with laser sensors "on the order of several thousand dollars apiece." He sees hope in this area though, with cheaper options popping up, such as Mobileye's vision-based localization system.
  2. Dealing with complex situations that arise only occasionally or in dense and uncertain (e.g., urban) environments. Dolan also said there's a "huge number of situations that can come up when driving, especially in urban environments, and the difficulty of creating algorithms capable of dealing with them flexibly and safely, as humans typically can."
  3. Performing adequate safety verification of the self-driving algorithms. This refers to the "impossibility of exhaustively testing complex software across all conceivable execution paths and environment inputs, leading to the need for simulation, statistical, learning-based or other intelligent means for performing safety verification and validation," said Dolan.

Still, Dolan thinks the outline, given enough time to accomplish, is feasible. He does raise a concern though; that when it comes to Musk's statement that Autopilot is safer than a human driver, the claim "isn't justified without qualification."

"'When used correctly' means that the human driver must pay full attention to the Autopilot with his hands on the wheel," said Dolan. "First, that requires a level of attention that may be significantly greater than that exercised by the average driver not using a Tesla Autopilot. Second, the population of individuals who own Teslas and use Autopilots is certainly different from the general population in many respects, one of which may include its level of care and attention when driving."

"It is still unclear how often, or drastically, drivers need to intervene in order to prevent unsafe driving behavior," said Dolan. "The availability of these data is a great thing, but they need to be analyzed more cautiously than by a one-to-one comparison with current human driving accident data."

Also see...

Visit TechRepublic