| Description |
This thesis describes the design, modeling, and gait control of a new bounding/rolling quadruped robot called the roll-U-ped. The robot has four uniquely-designed compliant legs for bounding gait locomotion, and the legs can reconfigure for passive and powered rolling. One of the main advantages of such a design is versatility as the robot can efficiently and quickly traverse over flat and downhill terrain via rolling and then transition to running for traveling over more complex terrain with a bounding gait. The contributions of this work are: (1) a detailed description of the robot design, (2) modeling and simulation of bounding motion, (3) investigation of bounding gait effectiveness using sinusoidal control inputs and inputs obtained from machine learning, and (4) prototype development and performance evaluation. Specifically, the prototype robot utilizes 3D-printed compliant legs for dynamic running and rolling, and the dual-purpose leg design minimizes the number of joints. Two functional prototypes are developed with on-board embedded electronics and a single-board computer running the Robot Operating System for motion control and evaluation. Simulations of the bounding gait locomotion are shown and compared to the performance of the prototype designs. Additionally, the robot's running motion is investigated for two types of inputs: a sinusoidal trajectory and a learned gait using the Q-learning technique, where results demonstrate effective running and rolling behavior. For example, using sinusoidal inputs, the robot can run with a bounding gait over a flat and stiff sandpaper-like surface at speeds of up to 0.21 m/s. On the other hand, over a flat and tacky-cushioned surface, the speed is measured at 0.14 m/s. Simulation results for Q-learning show gait speeds of 0.22 m/s for the tacky-cushioned surface, where experiments on the physical system yielded a gait speed of 0.15 m/s. For powered rolling, the robot was able to reach a speed of 0.53 m/s over a flat-smooth surface. The results demonstrate proof-of-concept of the design and feasibility of using machine learning to determine inputs for effective running locomotion. Finally, possible future improvements to the design, modeling, and motion control of the robot are discussed. |
