Multi-sensory autonomous mapping and exploration with application to robotic object rearrangement

The ultimate aim of this thesis is to develop an autonomous exploring and mapping mobile robot system to provide a platform for useful applications such as autonomous cleaning or tidying. The first requirement for an autonomous mobile robot is autonomy of navigation which can be assisted by implementing Simultaneous Localisation and Mapping (SLAM). SLAM enables the robot to incrementally build a map of the environment and localise itself with the map. Unfortunately, SLAM does not encompass motion control that enables the robot to move autonomously to build the map. Therefore, in order to achieve a truly autonomous mapping operation, this thesis also focused on the development of an autonomous exploration approach. In addition, this thesis also included an autonomous online map merging system which allowed the robot to perform intermittent exploration and yet still be able to acquire a globally consistent map. Finally, the robot system with autonomous mapping and exploration capabilities was extended to perform an autonomous object rearrangement. Particularly, this thesis concentrated on the development of a system for the robot to autonomously discover and restore changes in object positions without the assistance of external aids. This thesis made the following original contributions to research. Firstly, a time synchronisation approach was developed, which permitted high speed mobile robot mapping by minimising time synchronisation errors of odometry and range sensors. Secondly, a real-time system for robust and accurate SLAM with fusion of advanced sonar features and laser Polar Scan Matching (PSM) was presented. This resolved the problem of map drifts in corridors resulting from ill-conditioned laser scan matches. In the development of the SLAM system, the PSM was extended for multiple laser rangefinders. This solved the problem of scans overlapping insufficiently in the matching process given the limited field of view of a single laser. Moreover, a method of simultaneously calibrating odometry and fine tuning the local range sensors’ pose relative to the robot origin was presented. Thirdly, a Voronoi-loop-based approach for autonomous exploration which facilitated truly autonomous exploration and stable mapping by performing frequent loop closing was presented. Fourthly, an autonomous online map merging system for a single robot performing intermittent explorations was introduced. The map merging was achieved by combining a probabilistic Haar-based place recognition system using omnidirectional vision with an integrated SLAM and autonomous exploration system based on laser scan matching. This system was shown to robustly and autonomously perform map merging in challenging indoor environments. Fifthly, a method that utilised SLAM and autonomous exploration to discover and restore changes in object positions was presented. In the development of the object rearrangement system, a novel electromagnetic based grasping system and a novel path planning technique for transporting objects were introduced. The technical contributions of this thesis were validated with a series of practical experimentations using a real-time robot system with various combinations of sensors in typical indoor environments