A Framework for Identifying and Simulating Worst-Case Animal-Vehicle Interactions

摘要

This paper describes the development of a framework for identifying potentially dangerous interactions between intelligent vehicles and deer on public roadways. This framework is meant to be a flexible tool for improving simulation-based design and testing of collision avoidance systems for intelligent vehicles by providing more realistic test cases. Using a genetic algorithm and a biologically inspired parameter space, simulated deer were trained to be difficult to avoid for four rudimentary simulated drivers exhibiting prototypical responses to the deer's flight across the road. “Fitness” of a population of simulated deer was evaluated using a deer's minimum distance to a simulated vehicle during a roadway crossing event. Results of preliminary simulations show promise for the use of the framework in evaluating the effectiveness of more sophisticated human and/or robotic drivers.