Mixed-Neighborhood, Multi-speed Cellular Automata for Safety-Aware Pedestrian Prediction

Sebastian vom Dorff, Chih-Hong Cheng, Hasan Esen, Martin Fränzle
Proceedings, volume 13085 of Lecture Notes in Computer Science
19 th International Conference on Software Engineering and Formal Methods
Predicting pedestrian movement in unregulated traffic areas, such as parking grounds, marks a complex challenge in safety for automated vehicles. Without the ability to make certifiable predictions and judgments about safe interactions with other traffic agents in a real-time capable and economical fashion, the goal of self-driving vehicles cannot be reached. We propose a computationally efficient model for pedestrian behavior prediction on a short finite time horizon to ensure safety in automated driving. The model is based on a cellular automaton, working on an occupancy grid map and assumes a physical pedestrian capability constraint. It is enriched by a variable update rate with a mixed neighborhood, overcoming the limitations of vanilla cellular automata and coming closer to the results of state-of-the-art algorithms, while keeping the benefits of its straightforward parallelizability. The approach is evaluated on …
12 / 2021
Springer, Cham