Cooperative and Intrinsically-Correct Control of Vehicles in Diverse Environments (CoInCiDE)

点击翻译按钮获取中文摘要

While advancing assisted driving for single cars to automated driving has been put on the development agenda of many car manufacturers recently, the common understanding of experts is that the anticipated advantages of automated driving will fully unfold only if vehicle cooperation is considered: The enhancement of traffic flow, the avoidance of traffic jams and accidents, as well as relieving passengers from the task of supervision does require interaction and cooperation between automated and non-automated traffic participants. This requirement poses two main challenges: (i) to determine an agreed set of driving plans for the involved cars timely, such that the driving goals are met and the environment is appropriately considered, (ii) to guarantee safety by excluding behaviors possibly leading to collisions with any traffic participant, and, in particular, to exploit cooperation to predict and avoid situations that are potentially dangerous. This project targets exactly these two challenges: In the first phase, we have developed a hierarchical architecture combining behavior prediction, computing safe driving options, and establishing cooperative driving strategies by using principles of auction theory. The success of this solution concept has been demonstrated for a selected number of scenarios by simulation-based evaluation. The tests have also lead to the insight, however, which further developments are necessary for realizing an intrinsically safe and at the same time practically sound solution to cooperative automated driving. The second phase (for which funding is requested by this proposal) will in particular target the open issues of: (a) guaranteeing that cooperative driving strategies are timely available when required, i.e., the offline and online parts of plan generation are interlaced such that a safe driving strategy is always available; (b) accounting for the mismatch between the models used for planning and the true vehicle dynamics by establishing model conformance and robustness to uncertainty, (c) considering that the automated vehicles may use different concepts and metrics for cooperation, i.e., we define a common basis for vehicle agreement over different cooperation mechanisms; and (d) testing the developed architecture not only by simulation but on real automated vehicles; hereto, challenging scenarios will be determined, and an implementation of our planning and control hierarchy will be evaluated on DLR's three fully automated vehicles. The respective software for behavior prediction and cooperative planning, as well as the interfaces to the automated vehicles will be made available to the complete consortium involved in the SPP 1835.