A coevolution mechanism derived from competition relationships between ecological species is merged into Particle filters (PF). The new version of particle filters is termed Coevolutionbased adaptive particle filters (CEAPF). In CEAPF, samples are clustered into species, each of which represents a hypothesis of state of the system in a higher level than a single sample. Since the coevolution between the species ensures that the multiple distinct hypotheses can be tracked stably, the problem of premature convergence of PF can be solved. And the number of samples can be adjusted adaptively over time according to the uncertainty of the state of the system by using the population growth model. In addition, by using the crossover and mutation operators in evolutionary computation, intra-species evolution can drive the samples move towards the regions where the desired posterior density is large. So a small number of samples can represent the desired density well enough. And CEAPF is applied to robot localization in highly symmetric environments. Experiments prove that CEAPF can considerably improve the success rate and precision of localization.
展开▼
