Environment representations for automated on-road vehicles

One of the key challenges of any Automated Driving (AD) system lies in the perception and representation of the driving environment. Data from a multitude of different information sources such as various vehicle environment sensors, external communication interfaces, and digital maps must be adequately combined to one consistent Comprehensive Environment Model (CEM) that acts as a generic abstraction layer for the driving functions. This overview article summarizes and discusses different approaches in this area with a focus on metric representations of static and dynamic driving environments for on-road AD systems. Feature maps, parametric free space maps, interval maps, occupancy grid maps, elevation maps, the stixel world, multi-level surface maps, voxel grids, meshes, and raw sensor data models are presented and compared in this regard.