A Control Architecture for Dynamic Execution of Robot Tasks Trained in Real-Time Using Particle Filters

Abstract

With the advancement of technology and development of new gadgets to make life easier, the need in our society for task automation has increased over the years. Simple electronic devices, such as automatic vacuum cleaners, help to accomplish basic automated tasks. Robots built to complete more complex tasks are expensive, difficult to train, and don't normally exist in the home environment. Advances in human-robot interaction concepts and technology, along with new methods for training robots to complete tasks, have aided the development of new systems that are more practical to common users. In this thesis, an on-line training system for behaviors is implemented in real-time using a particle filter which allows for continuation of training if testing does not yield favorable results. Also a control architecture is designed and implemented to allow for execution of task sequences. This allows for larger, complicated tasks to be completed through the execution of smaller, simple tasks. This new training implementation is demonstrated and proposed for use in future training applications.