NRI: FND: Collaborative Control for Wearable Robots

点击翻译按钮获取中文摘要

There are many applications, including manufacturing, assembly, health care, and construction, where workers may be hampered by not having enough hands to do their job effectively. While one approach is to have a mobile robot assist the human, our project instead focuses on the augmenting human capabilities by developing a wearable robotic arm. Such augmentation of the human body can enhance a person's power, efficiency, safety, and quality of work. The project aims to make these wearable robots act as collaborative teammates, rather than directly controlled passive tools, making them both intuitive for novices and adaptable to expert users. This will significantly improve the efficiency and acceptance of such robots, and positively affect the work conditions of people interacting with them. The robot arm will be tested with human users performing tasks for which a third arm is useful. Whereas wearable robotics is a maturing field, there is almost no research on the human-robot interaction (HRI) aspects of such robots. This project investigates collaborative HRI for wearable robots, in four phases: 1) Collecting wearable collaboration data with a human and a tele-operated robot arm; 2) using this data to design an anticipatory Conditional Random Field (CRF) model for the collaboration; 3) developing a Partially Observable Markov Decision Process (POMDP) controller for the robot; and 4) evaluating the controller in human-robot interaction experiments with a physical wearable robotic arm, using metrics of efficiency, fluency, and usability. By doing so, the project contributes to the state-of-the-art in computational HRI by developing new probabilistic models for human-wearable robotic collaboration. The project also contributes new empirical data on how people interact with wearable co-robots through two human-subject studies. The collected data set on human-wearable-robotic interaction will be released to serve the research community.