A Bayesian Approach to Context-based Recognition of Human Intention for Context-Adaptive Robot Assistance in Space Missions

Abstract :

Long-duration space exploration (LDSE) missions with a high demand for intelligent and adaptable robotic assistance are expected to start in the next decades. Robot assistants deployed on these missions should be able to interpret the intentions of astronaut(s) and provide assistance accordingly, under continuously changing interaction context. This necessitates the development of robust and context-adaptive models for inferring human intentions. However, current state-of-the-art artificial intelligence (AI) algorithms rely on expensive data-driven machine learning (ML) models that provide inferences assuming a set of known contexts. In this paper, we present a context-adaptive hybrid architecture for intelligent and adaptable Human Computer Interaction (HCI). This architecture uses state-of-the-art machine learning models for the recognition of specific contexts within the robot’s environment, and a two-layer Bayesian network to infer human intentions based on the current context. The Bayesian network enables the integration of human expert knowledge about the mission goals within the inference process and can also deal with missing or uncertain information about context. Thus, the proposed framework increases robustness and reduces the need for large and complex (high dimensional) human-robot interaction data, which is unavailable in space applications. Context-adaptive interpretation of human intention would promote an intuitive interaction with robots, which in turn could reduce the overall stress in astronauts while interacting with robot assistants in isolated, confined or extreme (ICE) conditions.

Keywords :

Bayesian networks, intention recognition, context, HCI, HRI, HMI

Files:

lubitz_kimmi_cobabir_2022_-_Adrian_Lubitz.pdf