A multiple vehicle group modelling and computation framework for guidance of an autonomous road vehicle

Predictive autonomous vehicle guidance schemes can be configured to embed human driver-like decisions regarding multiple dynamic obstacle vehicle groups prevalent in public traffic, especially on highways. This paper proposes a vehicle grouping model and computation algorithm that facilitates dynamic grouping of surrounding object vehicles. This grouping serves to compute the time varying areas that are to be occupied by vehicle groups in the predicted motion plan so that those areas including the undesired local minimums can be excluded for the non-convex motion planning problem. To reduce the computational burden of the grouping and boundary generation for online implementation, supervised learning methods are applied to train the neural networks that compute the optimal boundary of the group (s) that maximizes the available planning field. The proposed algorithm is incorporated in a predictive guidance scheme and its performance and computational details are illustrated via simulations of an autonomously controlled vehicle in public highway traffic scenarios involving multiple other object vehicles.