What is the real Cost of Driverless Cars at what level of functionality?

Tracking the costs of driverless car technology is not easy. Almost everyone keeps their preferred technology secret, and the costs are coming down really dramatically in announcements (but perhaps not in delivery.) At the moment (Dec. 2017) no fully autonomous cars (let’s call them autos) are available for mass production, but we are close. At the high end, (Tesla, Audi, Mercedes, Cadillac) offer some appealing ADAS (automatic driver assist packages) like Super Cruise and Autopilot, but they are all dumbed down systems of what could be done. They all argue that these dumbies are available for 6 to 7 thousand dollars. This is in stark contrast to really functional systems with powerful computers, LiDAR, radar, sonar, integrated circuits, cameras, AI devices, GPUs like Nvidia’s Tesla and Titan at thousands each, that together still cost about $50,000 or more. And that does not include the cost of the software, however that is delivered.

At the moment, only the Audi A8 even uses Lidar. Tesla has sworn not to use it and even broke up with Mobileye over petty differences about this. Many prognosticators believe Musk will eat his words as cheaper and more powerful Lidar with no mechanical moving parts come on the market, but in spite of many announcements by Quanergy, Ouster, Valeo, and even the granddaddy of them all, Velodyne; they are not in delivery products yet. In fact, the production leader is still Velodyne with a traditional LiDAR with 128 lasers at $40,000 a pop.

So, the near future is clear: really functional driverless cars will be restricted to the very high end for individual driver owners. Mobility services are another matter where even very expensive driverless systems will soon pay for themselves from reduced accidents and the cost of a human driver. This is the real meaty part of car manufacturers’ plans: becoming mobility services. An Intel study decided that total income for car manufacturers will rise from $3 trillion to $7 trillion if they become mobility services. So that is their goal.

At the low end, crummy autos will be developed that reduce accidents a tiny bit from their current high rate but provide nowhere near the safety margins on high end autos. The prototype for this is the Nissan 2018 Leaf, that may even have shadow drivers monitor and take over autos in difficult patches where visibility is bad or road markings are not up to scratch.

In 2012 Google’s driverless test cars had about $150,000 in equipment including a $70,000 lidar (laser radar) system which made it too expensive for consumers.

By 2016 the bad news was: You couldn’t buy a fully autonomous car yet. Such vehicles were still only allowed on the roads for testing purposes. And chances are, you couldn’t afford one yet anyway. Adding driverless technology to a car would have increased its cost by $70,000 to $100,000 in 2016.

But the good news was that every car manufacturer was jumping headlong into development. They had realized that mobility services with autos was their only future.

IHS Automotive predicted that 10% of new cars sold worldwide in 2035 would be fully driverless (as in, they wouldn’t need any human assistance) and that the technology would become ubiquitous sometime after 2050. They will soon eat those ultra conservative words.

Another popular target for most ADAS developers is the sensor array. Tesla’s Elon Musk believes the LIDAR/RADAR sensor approach is too expensive. “It’s better to have an optical system, basically cameras with software that is able to figure out what’s going on just by looking at things,” he recently told Bloomberg. This seems very silly with lidars plummeting in price while functionality improves to cover near, medium, and far fields in one lidar.

Delphi Automotive Plc, which changed its name to Aptiv Inc. after buying nuTonomy for half a billion dollars, says it will cut the cost of self-driving cars by more than 90 percent to around $5,000 by 2025, according to Chief Executive Officer Kevin Clark. This is pretty much an empty statement since it does not mention safety, and you can have crummy autos for $2,000 and really functional autos that virtually eliminate accidents for $100,000.

“Just a few years ago, a single top-of-the-range lidar cost upwards of $75,000. Today, we’ve brought down that cost by more than 90%,” Krafcik said. “As we look to scale, we will do even better, with the goal of making this technology accessible to millions of people.” Again, this makes no mention of whether the lidar is has one laser or many; is short, medium, or long range, what its resolution is; or a host of other important complicating factors.

Since its collaboration with Waymo, Velodyne has to range from $8,000 to $30,000, depending on how many lasers it shoots out. Ford and China-based internet company Baidu have both invested $150 million in Velodyne for their self-driving car efforts.

Let’s go over some of the other relevant specs for a lidar. Frame rate is usually a big deal for LIDAR systems that spin, but with solid state lidars , you just get a million points per second and you can decide what frame rate you want to deal with, since it’s all software controlled. Field of view is generally about 120 degrees both horizontally and vertically. The minimum range is about 10 centimeters, and the maximum range is at least 150 meters at 8 percent reflectivity. At 100 meters, the distance accuracy is +/- 5 cm, and the minimum spot size is just 9 cm. A solid state lidar is a small puck, 9 cm x 6 cm x 6 cm. Produced in volume, a unit solid state lidar will cost $250 or less. Volume may be about 100,000 units, which isn’t a lot in the automotive space.

Redundancy is much needed, so one lidar will not be enough for a high end system. As an example of high automation, the airline industry uses triple redundancy and 3 independent processors with code written by 3 different teams in 3 different languages.

In comparison to this, current systems in production are barely ADAS and certainly not autos.

The Audi A8 seems to have the best of current production semi-autonomous systems.

David Silver seems to think Cadillac’s system is the best

But I really don’t agree.

The Audi A8 beats other high end systems. Mercedes-Benz sounds an alarm if you take your hand off the wheel. Tesla makes you hit the turn signal to show you want to change lanes. The new Cadillac CT6 goes so far as to monitor your eye movements with infrared cameras to make sure you’re paying attention to the road. That’s an emphatic statement of the car’s status: Level 2.

To be sure, the A8, like the Nissan Leaf, is designed only for slow speeds and traffic jams. It monitors the driver, even while in a traffic jam, and continues to do so as the speed edges up over the limit. If the driver falls asleep, it’ll wake him up; if it can’t get his attention, it will stop the car.

Pictured: Velodyne Puck, an existing lidar system

Joe is a bricoleur, trying to understand the complexity of the place of values in a world of facts, using only common sense.