Abstract

This review surveys contemporary methods for inferring human intentions and predicting human actions in human-robot collaboration, an important precursor for robots to provide useful assistance. The survey finds that intentions and goals are often inferred via Bayesian posterior estimation and Markov decision processes that model internal human states as unobserved variables, or that represent both agents in a shared probabilistic framework. An alternative approach uses neural networks and other supervised learning methods to directly map observable outcomes to intentions and to predict future human activity from past observations.