Evidential- and Particle-Based Drivable Dynamic Occupancy Grid

Comprehensive modeling of the surrounding environment is essential for autonomous vehicles. Occupancy Grids are widely used for this purpose due to their ability to represent any type of obstacles, as well as free and unknown space. However, in addition to the occupancy state, additional features are required to effectively describe driving scenarios, including obstacle dynamics, obstacle classification, and drivable space. This work proposes a novel Drivable Dynamic Occupancy Grid that jointly estimates these four features. This objective is accomplished through a temporal filtering that employs an evidential representation to jointly model occupancy and semantic data, in combination with a particle filter for estimating dynamic states. The obtained unified representation ensures a consistent and homogeneous scene description, while also enabling the harnessing of synergies between the estimated features to enhance overall estimation. Experimental results using real sensor data from urban scenarios demonstrate the effectiveness of the proposed method and show improvements over previous work.