Socially-inspired fully decentralized robot coordination

Abstract

The necessity of the collision-free determination of movement trajectory can be observed among both animals and people. In a completely autonomous way, unrelated units make decisions on choosing their movement trajectory while moving in a common space. There is no explicit communication between the individual participants of the movement, and there is no central planner coordinating their activities. Nevertheless, large groups of unrelated movement participants are able to coordinate their actions, taking the trajectories of other road users into account and performing the task that is entrusted to them. These observations have inspired us to create a decentralized algorithm for coordinating the movement of mobile robots based on a model of the behavior of moving people. The need for the autonomous coordination of many objects and their management sets high requirements in terms of stability, scalability, and flexibility. The use of simple reactive controllers is a remedy that meets the limitations that are imposed on the algorithm. The solution provides methods for defining a collision-free path of movement based on the current observation of a robot’s environment. The behaviors that are modeled in the controller have been inspired by natural phenomena such as respect for those who stand higher in the social hierarchy, giving priority to people leaving a room, or generally known traffic rules. This paper presents algorithms for mobile robot coordination that are inspired by such phenomena. The algorithms were tested in both real and virtual environments in several settings, such as open space, a narrow passage, and a narrow corridor, and the results are presented. The experimental results were compared with the results of the recognized reciprocal velocity obstacles (RVO) (Van den Berg et al., 2008) method. Moreover, additional video materials are available online; these show the real-life robots in action when running the proposed route-planning algorithms.