Autonomous exploration and mapping of complex 3D environments by means of a 6DOF manipulator
- Publication Type:
- Issue Date:
The futuristic vision of industrial robotic systems that operate in complex, unstructured and diverse environments is beginning to become a reality due to the advances in computing, sensing and control. Automatically acquiring the structure and the properties of an environment in a timely manner is one of the key tasks that need to be accomplished in many field robotics applications. This thesis presents a novel and efficient approach to the exploration of three-dimensional (3D) environments using an industrial robot manipulator. The approach presented combines the objectives of 3D map building and surface material- type identifcation. The manipulator is manoeuvred through a sequence of viewpoints that are selected to maximise the quality of the map generated, minimise the time taken for the exploration, as well as minimise the uncertainty of the surface material type estimation, all whilst avoiding potential collisions between the manipulator and the environment. The thesis first focuses on acquiring the geometry of surfaces in the environment while exploring the industrial robot manipulator's collision-free configuration space. Ellipsoidal virtual bounding fields are positioned around the manipulator's links so that distance queries can be performed and collisions with obstacles in the environment or unexplored space are avoided. Information theory is used to measure the information remaining on the geometric map and the manipulator's configuration space. A sampling strategy is used to select candidate viewpoints which are predicted to reduce the information remaining to measure. Each viewpoint enables the manipulator to position and orientate a sensor so that environment data can be gathered. The candidate viewpoint solutions can then be ranked based upon the exploration objectives. The collected sensor data is fused into a map. The map is then segmented into groups of Scale-Like Discs (SLDs), which are generated via principal component analysis. Once the surface geometry becomes available, a strategy is required to maximise the accuracy of the surface material-type identification. Surface material-type identification is made possible through intensity measurements, which indicate the refl ectivity of the surface when illuminated by an infra-red laser. Thus, identification is significantly infl uenced by the relative geometry between the sensor and the surface to be identified. Information theory is used again to determine surfaces which have not had their surface material- type identified. Appropriate viewpoints facilitating accurate identification are selected by solving an optimisation problem using the Levenberg-Marquardt algorithm. This two-stage exploration approach is shown to successfully determine viewpoints enabling an accurate environmental map to be generated. The proposed algorithms and approaches are integrated into the system, Autonomous eXploration to Build A Map (AXBAM). Extensive experimental studies have been conducted on a complex steel bridge structure using a Denso industrial robot that has been equipped with a laser range finding sensor. These experimental studies demonstrate the efficacy of the AXBAM system.
Please use this identifier to cite or link to this item: