Automotive

Self-driving Fords are keeping their cool in the snow

View 5 Images
Ford says that an area that is covered in snow can be navigated by an autonomous car if the area has been driven and mapped by LiDAR previously
Ford
Ford says that an area that is covered in snow can be navigated by an autonomous car if the area has been driven and mapped by LiDAR previously
Ford
Using multiple different sensors allows autonomous vehicles to build up a detailed view of their environment and the data collected is combined in a process called "sensor fusion"
Ford
LiDAR (or Light Detection And Ranging) is used by autonomous vehicles to scan the world around them and create a navigable 3D model
Ford
According to Ford, the LiDAR sensors used by its autonomous cars are able to scan 2.8 million laser points a second
Ford
Ford says that its autonomous vehicles produce and process over 600 GB of data every hour
Ford
View gallery - 5 images

Self-driving cars are developing apace and are ultimately expected to become safer than those driven by people, but first they will have to overcome their biggest nemesis: snow. It's slippery, unpredictable and obscures road markings. Ford has given some insight into how its autonomous cars will deal with the white stuff.

The main weapon in the autonomous car's arsenal against snow is the LiDAR sensor. LiDAR (or Light Detection And Ranging) is widely used by autonomous cars as a means of scanning the world around them to create a navigable 3D model.

Of course, for an area to be accurately mapped using LiDAR, it can't be covered in snow. Ford says that an area that is covered in snow, however, can be navigated if it has been mapped by LiDAR previously. An existing map of the road and surrounding infrastructure created when the weather was good can be used as a baseline to identify the car's position even when it is snowy.

According to Ford, the LiDAR sensors used by its autonomous cars are able to scan 2.8 million laser points a second and are so powerful that they can identify falling snowflakes and raindrops. While this allows them to create high-resolution 3D maps, it also means that snow or rain can be viewed as an obstacle by autonomous cars. In order to avoid this, Ford has developed an algorithm that identifies precipitation and filters it out of a car's "vision."

LiDAR (or Light Detection And Ranging) is used by autonomous vehicles to scan the world around them and create a navigable 3D model
Ford

In addition to LiDAR sensors, Ford's autonomous vehicles use cameras and radar to map and monitor the environment around them. Using multiple different sensors allows the vehicles to build up a more detailed view of their environment, with the data collected combined in a process called "sensor fusion."

As well as providing what Ford calls "robust 360-degree situational awareness," sensor fusion allows for a degree of redundancy. In the event that one of a vehicle's sensors is covered by ice, snow, grime or debris, the vehicle can still operate using the data provided by the other sensors. Ford suggests that, in the future, autonomous vehicles may be able to clean or defog sensors themselves.

These approaches to scanning and modelling their environments are said to allow Ford's autonomous vehicles to locate themselves to within 1 cm (0.4 in), compared to over 30 ft (9.1 m) as is the case with GPS, the carmaker says. They do, however, produce and process over 600 GB of data every hour, which Ford says is more data than the average person uses in 10 years on their smartphone.

Ford carries out its winter weather road testing in Michigan, including at Mcity, and claims to be the first automaker to publicly demonstrate autonomous vehicle operation in the snow.

The video below provides an overview of some the technologies used by Ford's autonomous vehicles for navigating in the snow.

Source: Ford

View gallery - 5 images
  • Facebook
  • Twitter
  • Flipboard
  • LinkedIn
6 comments
Bob Flint
Humans & their internal computer can easily process (depending on the age 10 to 100 times that data or more) but more importantly process it in split seconds. I don't need to know how many snow flakes are falling, I can see and judge the intensity as well as the standing water, and many other inputs. Perhaps the missing link would be the ability to improve night vision, and employ thermal imaging with an overlay on the windshield, rather than counting snowflakes..
Mel Tisdale
Just how much is one of these vehicles going to cost by the time they have covered all bases - assuming such a task is possible? (What happens when one of these vehicles approaches a sink hole?)
Bob has it right, develop these systems for driver assistance instead of for being fully autonomous and you immediately plug in a processor that years of evolution have honed to outperform anything we can make. (I have yet to see any performance figures for the commonsense level of even the most modern manmade computer.)
And please don't tell me that Ford is going it alone in developing the map their autonomous system is going to use. If ever there were a situation where vehicle manufactures needed to share their methodology it has to be in drawing a common map for all navigation requirements. This should then match the maps road planners hold on their computers.
McDesign
Mel - I think the answer is that soon, all cars will share both their observations and their learning with each other in real time - our individual thousands of hours of learning will happen concurrently and very quickly, distributed among some large number of observant cars for automobile AI.
Mark Twain's "Life on the Mississippi" talks about how much riverboat captains talked to each other at every opportunity, to keep them all updated about a river that could and would change from day-to-day.
Same thing, but far faster.
Douglas Bennett Rogers
Very valuable as an unmanned delivery vehicle, as only cargo is at risk in that vehicle. Also, doesn't experience bad weather stress.
Stephen N Russell
Need to test cars in No CA Sierra Nev Range which does get Heavy snow. Aside Midwest. & take 4X4 to Alaska.
Mel Tisdale
@ McDesign
There is more to the map that these vehicles require than can be achieved by social networking. Autonomous vehicles need to know when there is a diversion (I doubt that there will be onboard sensors that can read "DIVERSION" signs and suchlike). Also required will be road closures for parades, carnivals etc. with time, date and planned duration plus 'road now clear' information. On top of that will be a record on the map of any roadworks with lane closures etc. This surely calls for one map that is updated in real time 24/7. That in turn calls for the map to have one and only one source. (Just look at the mess the accountants have made of sat-nav and their failure to provide free map updates 24/7 as they change. There is probably a lot more that would come out of a combined attack on the problem.