Ford Working On Autonomous Driving In Snow Conditions

This post is also available in: עברית (Hebrew)

The recent collision between the Google Car and a Californian bus notwithstanding, autonomous vehicle generally manage to get around quite well. They “see” the world around them in many different angles, analyse it, and respond accordingly. That is, until snow arrives. Snow throws a spanner in the proverbial spokes, sends video analysis systems haywire, and prevents the cars from behaving as they should.

Most driverless car tests take place in sunny California or in controlled environments, where visibility is clear and nothing fazes the AI, but this is untenable if autonomous vehicles are ever to venture in the real world where snow and haze and sleet are not a rare occurrence. Ford has now revealed how it plans to tackle the problem.

“If self-driving cars are to become a reality, and they almost certainly will, they must be able to navigate snow-covered roads,” said the car maker in a statement.

To overcome this challenge Ford is using Lidar technology. Lidar uses lasers to accurately measure distances and to create a 3D maps from data collected in favourable weather conditions. These maps are then used in comparisons with data from other driverless vehicles to identify their positions when visibility is poor.

One issue that all manufacturers face is that Lidar often misidentifies snow and water particles as obstacles to be avoided. To this end, Ford is working with researchers from the University of Michigan to develop algorithms that can recognise the particles and filter out that information which is irrelevant for safety.

To bolster data reliability, Ford combines data from radar, cameras, GPS, and Lidar to monitor road conditions. At the same time, the sensors scan the environment for roadmarks, that, when found, are compared to the existing 3D map database. This gives positioning that Ford claims is more accurate than GPS. Ford calls this sensor fusion, and the fusion system amasses some 600GB of data an hour.

The diffusion of reliance on multiple sensors create for a more accurate and safer reading of the environment. If one sensor goes awry, the system as a whole should still function properly.