Adaptive rule of combination for observations over time

Landmark-based navigation of autonomous mobile robots require the integration of observations from sensors over time. This paper shows that Dempster's rule of combination is not appropriate for this domain. An alternative rule of combination for Shafer belief functions is derived, which adapts the belief updating process based on a contextual weighting parameter (n), where n is a function of the expected landmark permanence, change in sensor discriminability, expected potential for dynamic occlusions, and tracking error. These influences are fused using fuzzy rules. The adaptive rule is demonstrated empirically on a Denning-Branch MRV4 mobile robot to produce the desired behavior for navigating the robot to persistent landmarks for different values of n. The rule can also be used to estimate the interval of time that the robot can safely operate without a sensor update.