Safe navigation within dynamic environments: a partial motion planning approach

Abstract

In our work, we address the problem of autonomous navigation within dynamic environments partially known. This problem as motivated a lot of work over the last decade. However, we have noticed that some fundamental aspects of the problem, mainly related to the dynamic nature of the environment, have been put aside, to start with, the limited time a system has to plan and execute a trajectory, while placed within a dynamic environment. This real time constraint lies at the heart of our work. In this work, we present a new planning approach that explicitly accounts for the real-time constraint, imposed by the dynamic nature of the environment. Partial Motion Planning (PMP) is an original approach we propose in this work to address the problem. In fact, to our opinion, Partial Motion Planning is the best possible approach to the problem we observed, namely, the intrinsic incompatibility between the Motion Planning within a dynamic environment and the real-time constraint. Besides, PMP accounts for the safety issue, that we address from the perspective of the Inevitable Collision States (ICS) formalism. This concept provides strong safety guarantees to the system. Finally, in our work we demonstrate the efficiency of this approach PMP for relatively complex environments. Several experiments have been set up where PMP has been implemented on a real robotic platform, the Cycab, an electrical vehicle developed by INRIA.