Date of Award


Degree Type


Degree Name

Doctor of Philosophy (PhD)

Graduate Group

Electrical & Systems Engineering

First Advisor

Daniel E. Koditschek

Second Advisor

Cynthia Sung


Most models of legged locomotion assume a rigid ground contact, but this is not a reasonable assumption for robots in unstructured, outdoor environments, and especially not for field robots in dry desert environments. Locomotion on sand, a highly dissipative substrate, presents the additional challenge of a high energetic cost of transport. Many legged robots can be adapted for desert locomotion by simple morphological changes like increasing foot size or gearing down the motors. However, the Minitaur robot has direct-drive (no gearbox) legs which are sensitive enough to measure ground properties of interest to geoscientists, and its legs would lose their sensitivity if they were geared down or the footsize increased substantially. This thesis has two main contributions. First, a controller for jumping on sand with a direct-drive robot that saves significant energy in comparison to a nominal compression-extension Raibert-style controller without sacrificing jump height. This controller was developed by examining the complex interaction between the jumping leg and the ground, and devising a force to add to the leg controller which will push the robot’s foot into a more favorable state that does not transfer as much energy to the ground. The second contribution is a ground emulator robot which can be programmed to exert ground force functions of arbitrary shape. With the ground emulator, it is possible for a robot on a linear rail to jump dozens of times per experiment, whereas traditional experiments on granular media would require the ground to be reset between individual jumps. Results from the simulation experiments used to develop the controller and the ground emulator experiments used to test it on a physical robot leg are validated with experiments on a prepared granular media bed. Finally, the contributions of this thesis are contextualized in a broader project of building explainable artificially intelligent systems by composing robust, mostly reactive controllers.

Files over 3MB may be slow to open. For best results, right-click and select "save as..."

Included in

Robotics Commons