| This dissertation presents a simulation study toward full automation of a walking machine's navigation and guidance. In this dissertation, a navigation system, which includes the adaptive gait control module, the terrain data analysis module, and the local and global path planning module, is developed for an autonomous walking machine. The optimal quadruped crab gait is developed and the properties of the optimal crab gait are used for basic decision strategy in the adaptive gait controller, so that the adaptive gait can perform a nearly optimal periodic gait when the vehicle travels on level ground. The adaptive quadruped gait has a three level hierarchical structure: the body motion adaptation, the leg sequence adaptation, and the leg position adaptation. This type of hierarchical structure works well for the walking machine on the H-type terrain, and is suitable for six-legged machines too. The regional analysis techniques developed in the terrain data analysis module are general in purpose. Those algorithms provide useful tools not only in path planning but also in general 2D image analysis. The skeleton map, which combines both the concepts of network graphs and potential functions, has the information for global path planning. The properties of the r-symmetric axis extend the definition skeleton to more general and higher dimensional cases. From the necessary condition for r-symmetric points, it is found that omni-directional ranging sensor is helpful for navigating the vehicle in a dynamic but limited environment. The range scanner used in the walking robot provides a terrain elevation map for the vehicle in the foothold selection and also supplies the global terrain information by an address-content memory scheme. The structure of a rule-based system for local navigation in unknown terrain is quite flexible. The rule base offer an useful tool to solve the local path planning problems. |
