Incremental Task Model Updates from Demonstration

摘要: Fully specifying a task model such that a robot can perform a task in all situations and environments is intractable. Instead, we propose a novel algorithm that allows users to update a baseline model by providing demonstrations or corrections in the environment in which the robot operates. A set of model updates are proposed that make structural changes to a finite state automaton (FSA) representation of the task. These changes are instantiated through conversion into a state transition hidden Markov model (STARHMM). The STARHMM’s probabilistic properties are then used to perform approximate Bayesian model selection to choose the best model update, if any. We implement and evaluate the model selection component on a simulated block sorting domain. Initial results show that this formulation can choose models that sufficiently incorporate new demonstrations, while remaining as simple as possible.