Motion planning amongst arbitrarily moving unknown objects

An approach to motion planning amongst arbitrarily moving unknown objects is presented. As opposed to other approaches to motion planning we avoid the assumption that the motion parameters and the shape of moving objects are known a priori or can be predicted over longer time intervals. By giving up this assumption, traditional methods such as space-time representation and search in space-time no longer apply. Our approach is based on a massively parallel network of simple processing elements. A relaxation process, which is driven by the simultaneous execution of a simple formula in these processing elements, creates a two-dimensional distribution of real numbers, denoted as potentials, which encodes information about collision-free trajectories. Our approach is different from classical algorithmic motion planning in that we do not employ an analytical planning or search algorithm. Instead, desired behaviors, such as the avoidance of moving objects, are achieved through adroit manipulation of the two-dimensional potential distribution