Dynamic locomotion with a hexapod robot.

Legged vehicles offer superior mobility over natural terrain compared to traditional mobile platforms. Furthermore, their structural flexibility admits greater versatility in functionality. This thesis concerns the development of dynamically capable controllers for a hexapedal robot in order to achieve fast, agile and efficient locomotion. The first contribution towards this end is the design and construction of a hexapedal robot, RHex. The experimental results we present establish RHex as the first power autonomous robot to achieve speeds exceeding one body length per second over terrain approaching the complexity and diversity of the natural landscape. The combination of simple open-loop control algorithms and RHex's morphology exploits mechanical feedback to yield surprising energetic performance and robustness. Furthermore, the versatility of the design becomes evident in additional behaviors that we present, including turning and dynamical back-flips. The second contribution of this thesis is the development of high bandwidth state feedback controller for hexapedal locomotion. This approach lies in extreme opposition to our open-loop controllers and is inspired by the dynamical nature of running in animals. In particular, research in biomechanics demonstrates the descriptive utility of simple spring mass models across a large range of sizes and morphologies. Consequently, we adopt the well studied Spring-Loaded Inverted Pendulum (SLIP) model as a literal control target for a hexapedal alternating tripod gait. Now the design effort focuses on speed and agility, with the possibility that such high bandwidth sensor feedback may be rather costly to implement regarding both platform resources as well as runtime efficiency. Specifically, we introduce the idea of template based control , wherein we attempt to actively tune the natural dynamics of the robot to mimic those of SLIP. We use existing gait level SLIP controllers to achieve stable locomotion in simulation, with a simple and intuitive regulatory interface. Despite realistic actuation constraints, we identify stable limit cycles with large basins of attraction, significantly increasing the promise for an experimental implementation on RHex. Even though such an implementation is outside the scope of the thesis, only a few practical issues remain to be resolved before it becomes a reality.PhDApplied SciencesArtificial intelligenceComputer scienceMechanicsUniversity of Michigan, Horace H. Rackham School of Graduate Studieshttp://deepblue.lib.umich.edu/bitstream/2027.42/123265/2/3068951.pd