Inferring user intent with Bayesian inverse planning: Making sense of multi-UAS mission management

摘要

The goal of intent inference is to use observed behavior to predict underlying mental states and causal processes that are likely to have generated the behavior. A potentially powerful technique for inferring intent uses Bayesian inference in structured generative models for planning. We describe our adaptation of the Bayesian inverse planning framework to the multi-unmanned systems mission planning domain. We describe three experiments that elucidate the space of planning priorities in the domain, infer users' goals and priorities given their actions with a planning user interface, and predict users' next planning actions using inferences about their goals and priorities.