S. Carpin (Germany)
Holonomic motion planning, randomized algorithms
We propose a new technique for samples generation to be used in sampling based motion planners. Differently from almost all the previous literature in the field, we introduce correlation in the sampling distribution. Correlation is used to push samples into still unexplored regions of the config uration space. In order to record the distribution of the pre viously generated samples, a suitable data structure, called samples tree is introduced. The samples tree allows effi cient updates and searches, thus not resulting in a bottle neck for the overall performance. The proposed technique has been incorporated into the probabilistic roadmap plan ner algorithm and tested over a set of benchmarks. Numer ical results outline that the proposed technique seems to be a viable alternative to uniform sampling.
Important Links:
Go Back
