Abstract
Small wall climbing robots are useful because they can access difficult environments
which preclude the use of more traditional mobile robot configurations. This could
include an industrial plant or collapsed building which contains numerous obstacles
and enclosed spaces. These robots are very agile and they can move fully through
three dimensional (3D) space by attaching to nearby surfaces. For autonomous operation,
they need the ability to map their environment to allow navigation and motion
planning between footholds. This surface mapping must be performed onboard as
line-of-sight and wireless communication may not always be available.
As most of the methods used for robotic mapping and navigation were developed
for two dimensional usage, they do not scale well or generalise for 3D operation.
Wall climbing robots require a 3D map of nearby surfaces to facilitate navigation
between footholds. However, no suitable mapping method currently exists. A 3D
surface mapping methodology is presented in this thesis to meet this need.
The presented 3D mapping method is based on the fusion of range and vision
information in a novel fashion. Sparse scans from a laser range finder and a low
resolution camera are used, along with feature extraction, to significantly reduce
the computational cost. These features are then grouped together to act as a basis
for the surface fitting. Planar surfaces, with full uncertainty, are generated from
the grouped range features with the image features being used to generate planar
polygon boundaries. These surfaces are then merged together to build a 3D map
surrounding a particular foothold position.
Both experimental and simulated datasets are used to validate the presented
surface mapping method. The surface fitting error is satisfactory and within the
required tolerances of a wall climbing robot prototype. An analysis of the computational
cost, along with experimental runtime results, indicates that onboard real
time operation is also achievable. The presented surface mapping methodology will
therefore allow small wall climbing robots to generate real time 3D environmental
maps. This is an important step towards achieving autonomous operation.