| Multiple-robots cooperation of handling complex tasks in unstructured environments is not feasible at the current technical level. Using a human-robot team is more practical because of the human's intelligence and the robot's strength. Previous researches on Human-Robot Interactions were mainly focused on Compliant Motion and Reflexive Motion. The implementation of Compliant Motion by using a wrist force sensor is addressed in this dissertation. Moreover, further investigations on how to save human's effort with repeated tasks are considered in this research. Four real-time translation/rotation/scaling invariant trajectory identification and learning methods are developed for this purpose. They are Three Distances Method, Spherical Method, Cartesian Method and Batch Method. Strategies in saving computational time and/or memory storage, and dealing with inconsistency and/or disturbance are then discussed to make the trajectory learning methods practical. When the trajectory is determined, optimal load distribution is introduced to optimize the performance of Human-Robot Interactions. To help in designing, analyzing, and testing the interactive system, a number of models of human are simulated and tested. These models are Pre-Planning Model, Proportion-Derivative Model, Re-Planning Model and Simulation with Computer “Mouse”. Physical experiments, at last, are performed to illustrate and verify the analytical results. The biggest contribution of this research is that real-time translation/rotation/scaling invariant trajectory identification and learning methods are theoretically developed and experimentally verified, which can save human's effort with repeated tasked in Human-Robot interactions. |
