Master of Science

thesisThe reconfigurable structure of Articulated Wheeled Mobile Robots (AWMRs) permits them to traverse rough terrains. Predicting behavior of these robots on a rough surface is vital to carry out path planning and control. Several model-based approaches have been proposed in the past to predict this behavior, but the existing work has been mostly focused on soft soil, or hard terrains with a smoothly changing surface that are not realistic. In the future, AWMRs would likely be sent to more challenging scenarios that are extremely rocky, such as search and rescue sites, mining operations, and natural scree fields. However, there remains a need to focus research efforts on modeling AWMRs for these challenging scenarios. Hence, this thesis surveys the problems with state-of-the-art methods that make them fail on extremely rough-rocky terrains and proposes a new approach to modeling AWMRs that overcomes those shortcomings. The proposed method upgrades existing kineto-static motion models to consider multiple points of contacts for each of the wheels of the robot, which is vital for extremely challenging rough-rocky terrains. A formulation of the proposed method is presented that can be applied to AWMRs with multiple serial articulated chains, with or without branches, emanating from a main body link called the Platform. The formulation has been intended to be as general as possible, but one may have to update it to suit his or her AWMR. This formulation consists of a cylindrical wheel discretized into tread along the surface, a linear wheel-ground contact model, a multipoint differential kinematic model to find the total iv contact velocity at all the contact points, and a least-square solution to the equilibrium equation derived using kineto-static duality. It is shown how this formulation can be specialized to the COLE VII robot. Simulations are carried out that generate forward trajectories for the COLE VII robot on a smoothly undulating terrain and a challenging rough-rocky terrain. The proposed method is compared directly with existing state-of-the-art kineto-static model and Differential Algebraic Equation (DAE) based constraint dynamics model. The popular Open Dynamics Engine has been used as the benchmark to find the accuracy of the trajectories generated by the three techniques. Two-way Analysis of Variance (ANOVA2) is performed, which found no significant differences between techniques on a smoothly changing terrain, whereas the proposed method is significantly better than the other two techniques considered on a challenging rough-rocky terrain. The simulations also demonstrate improved computational efficiency over the DAE-based constraint dynamics model, which have already been proven faster than OpenDE. Thereby, it is proved that the proposed method performs as good as OpenDE (with an error 1.3%), but with an improved computational speed of at least 40%