Robotic exploration for mapping: Systems and algorithms

Robotic exploration can enhance the quality, efficiency and completeness of mapping for surveying in unknown, distant, and hazardous environments. In most cases, multiple position sensors are installed on the mobile robots for precise localization of mapping tasks. Usually, laser ranging devices on the robots generate highly detailed and dimensionally accurate coordinate data in the form of point clouds. 3D Digital Terrain Model (DTM), Computer Aided Design (CAD) drawings and other surveying or visualization products can be synthesized from the point cloud data. Laser scanning while in motion is subject to position sensing errors, and point clouds obtained from a single static vantage position are generally incomplete because no data points exist in occluded areas. Position sensors that have high precision at the high sampling rate required for 3D mapping are usually complex and expensive. Therefore, the map registration algorithm is desired to automatically to produce a high precision global point cloud map and work independently with positioning sensors. Meanwhile, a path planning algorithm based on an uncompleted map is desired for mapping robot travel efficiently through unknown environments and complete mapping tasks.;The work in this thesis is focused on building a mobile scanning system and algorithms required for map registration, occluded area recognition, and vantage position determination as path planning. A mobile robotic system and a 3D LIDAR scanner with video cameras have been developed as the platform for experimentation and algorithm validation. This system produces color rendered point clouds as maps. Multiple positioning sensors including GPS, IMU, steering potentiometer, and odometers installed on the robot provide rough positioning estimates for tracking the robot and validation. The color point clouds generated from the mobile mapping robot are used for verification and performance evaluation of the developed algorithms. The contribution in map registration algorithms includes two parts: color assisted automatic point cloud registration based on point surface normal orientation histogram and occupancy grid correlation; Hue Assisted Iterative Closest Point (HICP) algorithm for fine registration. The first part does not require any pre-known positioning information. The registration process is completed by solving the transformation correlation in Fourier domain. However, registration accuracy relies on the resolution of orientation histogram and occupancy grid, and therefore can only produce coarse registration output. The second part requires rough-aligned point clouds before registration starts, which are exactly the coarse registration products. The HICP iteratively registers point clouds paired wisely together. The results of this come with high accuracy, but take more time than the first step. Above two algorithms work together to register color point clouds during exploration together into a global frame.;The contribution in path planning contains a vantage position generating algorithm based on occlusion detection and frontier based exploration. Methodologies for identifying and selecting candidate vantage positions for mapping are discussed. For complete and efficient map generation, the map completeness evaluation is based on grid occupancy in 2D space and point cloud density in 3D space. The path planning algorithm for frontier best view is contained by map overlap ratio for the registration. A trajectory evaluation is performed during next best view point selection to generate the next vantage position to complete the exploration process.;This research has produced efficient color assisted automatic and iterative algorithms for registration of 3D point cloud map segments, and a path planning algorithm to generate vantage positions for robotic carriers explore unknown environments to complete the autonomous exploration and mapping task.