Distributed Path Planning for Robots in Dynamic Environments Using a Pervasive Embedded Network

Abstract

We investigate the application of a low-cost, pervasively distributed network to plan paths for mobile robots in environments with dynamic obstacles. We consider a heterogeneous system composed of small, embedded, immobile, possibly sensor-less, communication nodes and larger mobile robots equipped with sensors and manipulators. The embedded network serves as a pervasive communication and computation fabric, while the mobile robots provide sensing and actuation. The network is responsible for planning paths for the mobile robots even though paths are being created and destroyed dynamically. The embedded network provides nearly optimal path planning without the network nodes or the robots having global knowledge or localization capabilities. Path planning is one of the most fundamental and wellstudied problem in mobile robotics. Techniques such as D*[1] plan paths in dynamic environments where paths are created and destroyed. Inspired by these techniques, we have developed a technique for distributed path planning in environments were obstacles appear and disappear. Traditional path planning algorithms for dynamic environments typically require a single mobile robot to build a map, update the map as the environment changes, and then finally plan over the map. Instead, we use an embedded network distributed throughout the environment to approximate the path-planning space and use the network to compute the path in a distributed fashion. The algorithm essentially works as a distributed variant of the popular wave-front path planning algorithm, or a breadth-first search from the goal, propagating paths from the goal location. The embedded nodes make up the vertices of the path planning graph, and the network connections between them are the edges of the graph. Mobile robots can then use reactive navigation to traverse the graph by visiting the vertices (i.e. the embedded nodes) to the goal. In order to respond to changes in the environment this graph has to be maintained as edges are added and removed. For powerefficiency reasons we prefer algorithms that minimize com