A recursive Bayesian filter for landmark-based localisation of a wheelchair robot

An odometry model, represented by a set of nodes (waypoints), is considered to be the infrastructure of any probabilistic-based localisation method. Gaussian and nonparametric filters utilise an odometry model to localise robots, while predictions are made by the filters to actively correct the robot's location and coordination. In this work, we present a recursive Bayesian filter for landmark recognition, which is used to verify the pose of a robotic wheelchair at a certain node location. The Bayesian rule in the proposed method does not incorporate a control action to rectify the robot's pose (passive localisation). The filter approximates the robot's pose based on a feature extraction sensor model. Features are extracted from local environmental regions (landmarks), and each landmark is assigned with a distinct posterior probability (signature), at each node location. A node is verified by the robot when the covariance between the posterior and prior probability falls bellow a threshold. We tested the proposed method in an indoor environment where accurate localisation results have been obtained. The experimentation demonstrated the robustness of the filter to work for passive localisation.