Threat assessment for general road scenes using monte carlo sampling

Abstract

A stochastic threat assessment algorithm for general road scenes is presented. Vehicles behave in a manner which includes a desire to follow their intended paths comfortably and to avoid colliding with other objects. In particular, this can be used to detect indirect threats from objects that are not on a direct collision course, but may be forced into a collision course by the traffic situation. An example is when a vehicle has to swerve to avoid an obstacle and because of that the vehicle itself becomes a threat to another vehicle. The vehicles are on a direct collision course from the beginning, but the situation still poses a threat because of the obstacle. Control inputs of other vehicles are modelled as stochastic variables and the resulting statistical expressions are solved using Monte Carlo sampling. In any Monte Carlo method there is always a trade-off between accuracy, i.e., number of samples, and computational load. A further contribution of this work is a method to create denser sample sets without increasing computational load