NRI: Representing and Anticipating Actions in Human-Robot Collaborative Assembly Tasks

Georgia Tech Research Corporation, Atlanta GA

Investigators

Abstract

For robots to effectively collaborate with humans on a variety of tasks, they must go beyond responding to human actions to anticipating them. This is especially true in the domain of collaborative assembly tasks where the robot assists a human worker by providing tools or parts as required. In this project, a method for formally specifying collaborative assembly tasks is being developed that allows a robot both to understand the action of the human as well as to determine which action the robot has to perform and when. This project is making fundamental advances to enable task specification to be compiled or converted into a grammar-like description of the human activity. Given this description, a probabilistic inference method that integrates sensory information to analyze the actions of a human and predict which actions the human will take and when. The system learns the necessary perceptual detection information for human actions will be learned from small amounts of training examples of individual actions. By integrating these perceptual observations within a structured representation of the task derived from the specification, the robot can make effective predictions about the timing of human action and can thus anticipate when it will need to provide assistance and of what type. Given these probabilistic predictions the robot makes a plan of action that optimizes a collaboration measure such as how idle time of the human or overall task completion time. The broader impact of this project is along two dimensions. The first is within the small to medium enterprise manufacturing and assembly industry. Successful development of the technologies described is critical for human-robot collaboration in a variety of structured tasks in these domains. Second, the ability to successfully anticipate human behavior is essential for the general integration of assistive robotics into society.