Experimental study of data merging techniques for workspace modeling with uncertainty

The vast majority of sensors used in autonomous robotic systems are submitted to uncertainty sources that often generate contradictory data that must be interpreted in order to optimize the reliability of the information extracted from the models that are built from those measurements. When certainty occupancy maps are used to represent the workspace of a robot, the estimation of the uncertainty level becomes a critical issue as it must become an active part of the model. Numerous techniques such as the Bayesian theory, the Dempster-Shafer theory of evidence and fuzzy logic inference schemes have beet; proposed to achieve data fusion of uncertain measurements. However, the performance of these approaches has not been; extensively investigated and compared in the specific context of certainty occupancy maps construction. This paper presents the results of an experimental investigation that has been conducted to adapt, implement and evaluate these three data merging techniques to achieve smooth progressive refinement in the construction of occupancy grids based on cumulative uncertain range measurements. The context of the application considered is that of collision-free path planning for mobile robots.