This paper describes a mobile robot equipped with a sonar sensor array, Werrimbi, in a guided feature based map building task in an indoor environment. Common indoor landmarks such as planes, corners and edges are located and classified with a multiple transducer sensor array. Accurate odometry information is derived from a pair of narrow unloaded encoder wheels. Discrete sonar observations are incrementally merged into partial planes to produce a realistic representation of environment. Collinearity constraints among features are exploited to enhance state estimation. The map update utilises Julier-Uhlmann Kalman filter which improves the accuracy of covariance propagation through nonlinear equations and eliminates the need to derive Jacobian matrices. Correlation among map features and robot location are explicitly represented. Partial planes are also used to eliminate phantom targets caused by sonar specular reflection.

This paper presents the key steps involved in the design, calibration and error modelling of a low cost odometry system capable of achieving high accuracy dead-reckoning. A consistent error model for estimating position and orientation errors has been developed. Previous work on propagating odometry error covariance relies on incrementally updating the covariance matrix in small time steps. The approach taken here sums the noise theoretically over the entire path length to produce simple closed form expressions, allowing efficient covariance matrix updating after the completion of path segments. Systematic errors due to wheel radius and wheel base measurement were first calibrated with UMBmark test. Experimental results show that, despite its low cost, our system's performance, with regard to dead-reckoning accuracy, is comparable to some of the best reported odometry vehicle.

This paper presents an environmental acquisition strategy for a mobile robot using an advanced sonar sensor to achieve mapping navigation in an a priori unknown, imperfectly structured indoor environment. Most existing feature based strategies rely on unrealistic assumptions about the environment, while their grid based counterparts hinder localisation which leads to rapid degradation of map quality. A dual representation strategy is proposed here which exploits the strength of both a feature map and a grid map. With the advantage sensor, the environment is scanned and the obtained features are classified into planes, corners, edges and unknowns. The feature map is only updated with the first three types of features. The grid map is updated with all measurement, including the unknowns resulting from complicated objects, to enable obstacle avoidance. On the grid map, distance transform based exploratory path planning is implemented. Adaptation has been made so that an explore-local-first behaviour is exhibited.