Uncertainty modelling and motion planning of an inchworm robot navigating in complex structural environments
- Publication Type:
- Issue Date:
Many ferromagnetic structures require continuous inspection and maintenance routines to ensure longevity, structural integrity and aesthetics. For most structures, routines are performed by teams of personnel, with each individual performing specific tasks. These tasks may be highly hazardous; being performed at height, in confined spaces or in the presences of hazardous materials such as lead based paints and vehicle fumes. Adopting a robotic solution for inspections would significantly improve occupational health and safety for maintenance personnel, while increasing the quality and reducing the cost. An inchworm robot has been developed for inspection of confined spaces in the Sydney Harbour Bridge. With a 7 degree of freedom multi-link serial body and magnetic pads for adhesion, the inchworm robot provides a dexterous means for climbing and inspecting particularly difficult-to-access sections of the bridge. However, due to the structure and the adhesion mechanism of the inchworm type robot, deformation of the robot body (i.e. structural uncertainty) and inaccurate landing position of the permanent magnet adhesion pads (i.e. hand position uncertainty) cause imperfect knowledge about the robot state. This prevents safe motion in a real world setting. The combination of these uncertainties present a unique challenge in robot motion planning and collision avoidance which is not considered in the literature. This thesis first focuses on developing a model for representing the structural and hand position uncertainties. The model describes the uncertainty in the coordinate frame of reference for the joints. A 3D probabilistic force field (3D-PF²) algorithm is developed to incorporate the uncertainties and allow for smooth, collision-free path planning. A force field surrounds each link to prevent collisions with each force field sized to account for the dimensions of the link and the uncertainty at the joints related to the link. Force fields are used to generate repulsive forces which push the robot away from obstructions while an attractive force pulls the end-effector towards a goal location. A Line of Sight Tree (LoST) algorithm is developed for longer time-horizon motion planning with the 3D-PF² algorithm used for local motion planning. The LoST algorithm provides waypoints as goal locations for the 3D-PF² algorithm. Waypoints are found in a manner loosely based on the way a person views a scene whereby their gaze tends towards important regions such as the edges of objects. Extensive simulations and experiments have been conducted to test the performance of both the 3D-PF² and the LoST algorithms within a number of environments including the specific application scenario at the Sydney Harbour Bridge.
Please use this identifier to cite or link to this item: