Reactive Motion Planning for Mobile Robots

Abstract

The motion planning for non-holonomic robots in moving environments is a complex problem. The results obtained in the evaluation of the reactive lazy PRM method, proposed in this work, show the importance of finding a solution for this problem. In fact, the method's performance can be considered satisfactory if it presents a fast planning phase, reflex actions based on sensors that do not require expensive algorithms, an effective process of reconnection performed in milliseconds, and a process of re-planning that is executed if the Lazy PRM and DVZ' s parameters are appropriate. The planning time is reduced due to the incomplete collision detector whose work is complemented with the robot's sensors during the path execution. On the other hand, the assignation of direction angles to the nodes that conform the shortest paths obtained by the algorithm A*, produces curves that allow the algorithm to omit the optimization process (i.e., the smoothing process). With respect to the reconnection process, the paths obtained with the planner are conformed by a single Reeds & Shepp curve and based on the incomplete collision detector, making short the time and close to optimal the curves obtained with the algorithm. Since the reflex actions are provided by the DVZ method, it is possible to interrupt the reconnection and re-planning processes if necessary, without incurring in bigger problems. If the execution's parameters for the Lazy PRM and DVZ methods are adapted, the replanning process will not be called very often and will be successful in the absence of narrow passages. A reactive lazy PRM planner for dynamically changing environments is presented in this chapter. Although some promising results are shown in its present form, the planner could be improved in a number of important ways. This approach can be extended to use real robots and to solve the problem posed by small static obstacles. Besides, some cases where the reflex action was not sufficient to avoid collisions were observed during the evaluation tests. Theses cases are difficult because they require a more intelligent behavior in order to avoid the robot to be trapped. In those cases, it can be necessary to add a process that computes the trajectories of moving objects and corrects the path in real time. Finally, a very interesting topic in robotics is the study of non-structured environments. This methodology can be extended to solve these cases.