Map-building plans for cars

As the advent of autonomous cars edges closer, automakers are turning to map-building technology to enable driverless cars to navigate roads. HERE, a Berlin-based mapping company owned by BMW, Audi and Daimler, says its routes to build high-definition maps can accumulate 100 gigabytes or more of data, with the aim of keeping self-driving cars safe in all environments. Visual sensors can currently keep cars safely within lanes, as well as distinguish between solid and dotted lines indicating stop signs and exits. But John Ristevski, vice-president of reality capture for HERE, says that doesn’t go far enough. “If you’re on a five-lane freeway, you need to know which of those five lanes you’re in, which are safe to traverse and at what exact point that exit ramp is coming up”. While current sensors have an effective range of about 50 metres, the accuracy decreases in adverse conditions, including snow or rain, and when objects are obscured by vehicles ahead. At motorway speeds, the car can only “see” about a second and a half ahead. Driverless cars will also need to locate themselves exactly - being off by a few metres may place a car on the wrong side of the road. Commerical GPS systems are only accurate to around five metres, but this increases to 50 metres in urban canyons, and don’t work at all in tunnels. HERE is hoping to correct this through building several localisation layers into its maps, which can position a car within centimetres. The layers involve extracting features including bridges and road signs from images shot by the mapping vehicle, then comparing them to what the car sees through its cameras. However, Netherlands-based mapping company, TomTom, says this is too unreliable. Pieter Gillegot-Vergauwen, one of TomTom’s vice-presidents, says the company found that “trying to model reality down to every single bridge pillar and then triangulating it is too sensitive to change”. He adds that too many visual changes, such as a tree being cut down or the area altering between seasons, can throw off the accuracy. TomTom is instead creating a “depth map” using the mapping vehicle’s laser-surveying sensor systems. This continuously records distinctive shapes and distances in scenery, but doesn’t try to identify what each individual feature is. In this way, Gillegot-Vergauwen says by considering the whole stretch of road, the car can calculate its location if a tree grows or a truck is in the way.