Building and navigating a network of local minima

Abst ract We present a novel method that constructs and navigates a network of local minima of potential elds deened over multi-dimensional spaces. Though motivated by problems of motion planning for robotic manipulators, similar techniques have been proposed for use in other domains such as molecular chemistry and drug design. The method is based on building a roadmap of paths connecting local minima of a potential function. The novel approach consists of an uphill search strategy used to climb out of local minima and nd new nearby local minima, without doubling back on previous local minima. With this uphill search strategy, one can nd local minima otherwise diicult to encounter, and one can focus the search to speciic local minima and speciic directions from those local minima. The construction of the roadmap can be done in parallel with very little communication. We present extensive simulation results.