| Popis: |
This paper tries to develop a scheme for path planning which is adaptive to drastic changes in the environment, by modifying the scheme of the Structured Boltzmann Machine. The SBM is composed of two-layered network modules. We propose that paths planned in advance for different environments are learned in-dependently by such modules where they are represented by sets of sub-strategies. In a new environment, the modules are connected to merge their sub-strategies. By the combined effect of associative memory and the competition between modules, the sub-strategies for given states are extracted from the modules, and these sub-strategies make up a complete path for the new envi-ronment. The proposed framework is applied to a simplified path planning problem using a 2-link manipulator. |
