A new software system developed at the University of Michigan uses video game technology to help solve one of the most daunting hurdles facing self-driving and automated cars—the high cost of the laser scanners they use to determine their location.
I picked up on an interesting comment in the story about systems for self driving cars. Fundamentally it said, you cannot have a self driving car go where there is no map.
I have driven with car navigation systems that have low cost maps that are out of date, spatially inaccurate and don’t have things (as I’ve said before) such as speed zones. They also often don’t have information such as unpaved roads. I wonder how a driver-less car would know that you didn’t want to drive on dirt roads and therefore know where they are.
Updates would be crucially important. In my area they have just made some major changes such as moving the location of motorway on-ramps by a half mile or so. I wonder how a driver-less car with an old map would deal with that. Most people have maps that are out of date, after all change happens every day, it is hard for the best staffed and most conscientious mapping companies to stay in touch, especially when in between driving, they rely on central and local government to tell them of proposed and then completed changes.
I expect most driver-less cars to alert the driver of there is something outside that is different from what is displayed. I see lots of situations where it may not know how to interpret what is inconsistent with a map and suspect an appropriate solution would be something like the LIDAR Orthophotography systems mapping companies use.
So imagine this situation. An off-ramp has been moved as part of a road reconstruction process. You have a system which has map data that doesn’t show it. True situation I drove past frequently. There is a sign on the side of the highway saying the ramp has been moved, the new ramp is half a mile closer to you than it used to be. The original gantry sign is still showing the ramp to be where it used to be and the car map data-set still shows the ramp where it used to be. So two of the inputs show the existing ramp to be in the old location and it isn’t there. It may be that there are cameras that at 60 miles an hour can see the ramp, but it could also be an access road for the construction crew.
The computer decides to tell the human passengers that there is some inconsistency. Oh dear, this is one of those new driverless Mercedes where the front seats turn around so that the front passengers can turn around and chat to the rear passengers, that great feature that they have been raving about.
I’ll leave you to play out the scenarios. The safest one is that probably the car drives past the ramp for 4 miles, gets off at the next one, goes back on the highway another 3 miles to the ramp on the other side. The less safe ones are not so nice and could end quite differently.
One quick note that in fact prompted me to write this blog is that in traditional OEM car navigation systems there are 2 main areas where the cost goes up.
1. Each camp in the value chain gets a margin and margins on options are typically ‘parts’ margins which are many times higher than standard items in the car, where they try to keep the price of the drive away car low for competitive reasons. So the factory gets a margin when the nav is installed, international marketing gets a margin when it is shipped. The national office gets a margin when the car arrives and then the dealership obviously gets a margin as well for selling the extra feature. At parts rates, the nav system will frequently end up costing at least 4 times what the manufacturer received.
2. The maps. So many people discount the importance of up to date maps. Think about the cheap car navigation system (or even the top of the range system you own) and how often it is out of date. There are lots of reasons why this happens. One of the biggest is that navigation system manufacturers over the years have put a massive squeeze on the price they are prepared to pay for car navigation quality maps. In theory the argument was that it would be compensated for in volume sales, but often that hasn’t been the case. Ultimately that means that quality and currency suffer. You can’t build quality products on peanuts. That’s not a major when an intelligent human is driving, who can interpret the instructions when what they see is not the same as what’s in the map data.
So when you look at these cool ideas of turning the driver to face the back of the car and then expect them to react to an anomaly on the road ahead as they sip their coffee (I’m sure the driver of a driverless car will still not be able to consume alcohol, think about how safe it is to be able to go everywhere you currently drive.
Then think about how many errors there are in the car navigation map data with the system you are driving in. Also keep in mind that the data in the really expensive top of the range car nav system, is the same data that’s in the portable device, the difference is in the high tech gyroscopes, inertia sensors, cameras (not yet), wiring looms, interface to the ODBC2 port for data and so on, don’t think you are paying for a more sophisticated map.
This new system with cameras, based on gaming systems may improve on the scenarios I have described here. It is however complex, Many maps are quite accurate in urban areas, but things happen. A one way street is flipped in the opposite direction, ramps move, lanes are closed for maintenance, contra-flows are put on one way streets, all sorts of things happen every day. So tomorrow when you drive to work, use your navigation system and note everything you do that is different to what the map tells you and the think about being in a car that has to make those decisions.
Don’t get me wrong, I love the concept of a driverless car. It is already proven in controlled situations, for example in one of my blogs you can read about mines in Western Australia, where trucks are controlled by an operator hundreds of miles away and it works fine, But it is in a controlled situation.
We have a little way to go, methinks. If you look through the word cloud o my blog you will find lots more about car navigation, best devices and maps and more. I welcome your feedback and comments.