Bayesian Relevance for Enhanced Human-Robot Collaboration

Abstract

Intent prediction is a difficult yet critical component for seamless Human Robot Collaboration (HRC). As robots become increasingly involved in helping humans with a variety of tasks, ranging from part assembly to healthcare and more, it is crucial to model and understand human intention. Many works still do not take advantage of the inherent relationships between objects, task, and the human model. Current human intent prediction methods, such as Gaussian Mixture Models and Conditional Random Fields, are generally less interpretable due to their lack of causality between variables. A novel framework called Bayesian Relevance (BR) is presented for human intent prediction in HRC scenarios. The complexity of intent prediction is captured by modeling the correlation between human behavior convention and scene data. The proposed method leverages inferred intent predictions to optimize the robot’s response in real-time, ensuring smoother and more intuitive collaboration. In this paper, we use a Bayesian Network to predict human intent from a multi modality information framework. A demonstration of a HRC task, using a UR5 robot, exemplifies BR’s real-time human intent prediction and collision avoidance. Evaluations demonstrate that our multi-modality BR model predicts human intent within 2.69ms with a 36% increase in precision, a 60% increase in F1 Score, and an 85% increase in accuracy compared to its best baseline method.