A new approach of nonholonomic path planning for car-like robots
is presented. The main idea is similar to many existing
approaches which obtain a path in two phases. It is familiar in
nonholonomic planning that at first a holonomic path is planned
which is approximated by a nonholonomic one in a second step by
subdividing it into smaller parts and replacing them with local
paths fulfilling the kinematic constraints. These methods mostly
rely on probabilistic methods and heuristic optimization. Our
approach uses a holonomic preliminary path as well, but it
serves only as a "loose guidance" to the second phase of the
planning process. The final path is not required to contain any
of the intermediate points of the preliminary path at all. The
method is effective in environments consisting of narrow
corridors but having wider free areas as well which can be used
for maneuvering.
