Modular robot design and pose training.

摘要

For educational and search and rescue robotic applications, a modular robot system has been developed. The emphasis of this robot research project is to significantly increase the mobility of individual modules with a simple physical shape. This facilitates the task of attaching multiple modules into clusters to create a more complex and capable robot system.;The mechanical design, electrical design and methods of software development of the modular robot are described in detail in this thesis. The process of choosing the physical shape, required torque, and components are discussed. Mathematical modeling and FEA analysis are presented. Consideration is given to manufacturability and to ease of assembly. Several operating modes are demonstrated which are unique to this modular robot design.;During the testing period of this research, a new method of motion control programming, called "pose training" was developed. Pose training allows the user to teach the robot how to move by physically manipulating the module by hand. Poses are recorded and played back in succession to form motion control.;A simplified mathematical model was developed for the robot. This research lays the groundwork for a module to autonomously improve motions inputted through pose training. The software uses reinforcement learning to allow the user to control how much the robot gate is changed. This improves the robot gait without losing the original intent of the inputted motion.