Probabilistic Path Planning Techniques

The main objective of this component is to adapt and improve the current motion planning technology to MOLOG's aims. In their basic form, MOLOG's motion planning algorithms get a description of the environment, a description of the body to move, and a description of the kinematic constraints on the motion of the body. The motion planner produces a smooth path for the body satisfying the kinematic constraints, such that there are no collision with the environment. In addition we wish motion planners able to optimize certain criteria (like distance to dangerous areas, length of the path, or number of rotations) and to give the reason of potential failures (e.g., blocking obstacle identification).

Two recent techniques which basically perform probabilistic searches of the configuration space: the random motion planner (RPP) and the probabilistic path planner (PPP) will be the basis of MOLOG's planner. Both RPP and PPP have shown to be very powerful for solving various motion planning problems, but they are in no way ready-to-use packages. We will investigate the improvements and adaptations required to satisfy the specific needs of MOLOG (three-dimensional broad complex environments, varying kinematic constraints, `black box' in which there is no user interaction). We will address the problem of planning rigid motions, the extension of this basic problem to handle kinematically constrained motions and dynamic changes of the environment, and we will develop advanced tools to incorporate a notion of path quality via optimization algorithms. The failure analysis will be based on distance computations between the various connected components of the configuration space.