AutoNet2030 has developed and tested co-operative automated driving technology. One crucial point is the need to define what is exactly intended by cooperation of autonomous vehicles. Indeed, autonomy refers to the ability to take one’s decision alone and this is very important on road for safety and robustness. However cooperation refers to deliberation before decision, taking into account other users’ wishes and this is highly desirable for traffic efficiency (and social acceptance).
While many frameworks for cooperation look at perfect cooperation (i.e. without any autonomy for individual vehicles) either in centralized or distributed way, AutoNet2030 extended previous research to a hierarchical approach combining both aspects. The planning (medium-term) is cooperative while the control is kept local (i.e. the decision lies within the vehicle) ensuring safety decision remains to each vehicle’s intelligence but also a global cooperative behavior.
Technically, this is ensured by hierarchical MPC planning and control algorithms that adapts to both intersections and highway scenarios defined by AutoNet2030. Using advanced knowledge of the structure of solutions (using the topological notion of homotopy classes), MINES ParisTech implemented a fast implementation of the algorithms running in real time. This software was ported to Inria’s automated vehicles within the AutoNet2030 architecture.
This has been used to validate the architecture and the urban cooperative planning and control algorithms as defined in the AutoNet2030 work plan.