Bosch Unveils New Tech to Improve Accuracy of Self-Driving Cars
New "radar road signature" technology from Bosch will be used to create maps for high-accuracy self driving. The new map is the first to use radar signals for a localization layer.
Bosch's radar sensors have a detection range of up to 250 meters, compared to a maximum video range of 150 meters. It takes only about half as much data to send the same information from radar than it does from video; per kilometer, the radar signature only has to transmit five kilobytes of data. The map creation is more efficient as a result. The radar is also unaffected by weather and other conditions that can reduce visibility. Video-based maps can be unreliable when conditions obscure visibility and thus the relevant information.
Radar maps will use three layers: a localization layer, a planning layer, and a dynamic layer. The localization layer will use data to identify a vehicle's location in a lane and in relation to nearby objects, and will use a video localization map as well as radar. The planning layer is responsible for trajectory planning; it is used to determine individual maneuvers like moving from one lane to the next. It also has information about the road itself and about signs along it. The dynamic layer has information about conditions that can change quickly, such as traffic, construction, and parking.
Bosch and TomTom have been working to create the radar road signature since they began collaborating in July 2015. Bosch is known for radar sensors, and the two companies integrated the signature into the high-resolution overall map TomTom created. Existing radar sensors were modified to detect static as well as moving objects.
The maps will need a large fleets of vehicles collecting information. "We currently expect that we will need fleets for freeways in Europe, North America, and Asia Pacific that each consist of around one million vehicles in order to keep our high-resolution map up to date," said Dirk Hoheisel, a member of the Bosch board of management. Maps will be continuously updated with new information, particularly for the dynamic layer.