Constrained Initialisation of the Simultaneous Localisation and Mapping Algorithm

This paper presents a novel strategy for feature track initialisation in the Simultaneous Localisation and Mapping (SLAM) algorithm. Spurious observations and ambiguity in the data association during the initialisation phase of the algorithm can be resolved by initialising each observation as a new feature. When a number of observations are found to correspond to a single feature, a constraint can be applied to consolidate all of the observation information in a single estimate of the feature. The state vector can then be collapsed by removing the redundant feature estimates. Any observations that are not reconfirmed with additional observations can simply be removed from the state vector after some appropriate time period has elapsed. Application of the algorithm to data collected from an AUV are shown to demonstrate the performance of the approach.