Learning attributes for situational awareness in the landing of an autonomous airplane
Erik Blasch · 2002
The paper investigates situational learning, which utilizes mathematics of probability and evidential theory, to determine the perceivable importance of environmental cues as they contribute to situational awareness. The situation-awareness agent's goal is consistent with that of an aircraft pilot; namely, to land a plane under a variety of weather and runway conditions. Landing requires hypothesis selection which can be formulated as a situational-learning (SL) problem in which sensed states are represented as current situational beliefs. The objective of SL is to learn how to select the optimal set of mutually non-exclusive hypothesis in order to maximize the identification of the situation. Three methodologies for the combination of sensor measurements for situational learning are designed and analyzed for a system equipped with a position measuring device and identification sensors. Using a learning algorithm for searching, the a priori identification probabilities of recognition are known. The methods are: (1) recursive Bayesian where the probability of the current state is based on the a priori information multiplied by the likelihood function, (2) Dempster-Shafer(DS) which uses evidential reasoning/accrual to combine information of uncertainty, and (3) modified Dempster-Shafer(MDS) which uses a combination of evidential reasoning and probability analysis. The methods are assessed for cases with and without feedback.