Using Learned Action Models in Execution Monitoring

Planners reason with abstracted models of the behaviours they use to construct plans. When plans are turned into the instructions that drive an executive, the real behaviours interacting with the unpredictable uncertainties of the environment can lead to failure. One of the challenges for intelligent autonomy is to recognise when the actual execution of a behaviour has diverged so far from the expected behaviour that it can be considered to be a failure. In this paper we present further developments of the work described in (Fox et al. 2006), where models of behaviours were learned as Hidden Markov Models. Execution of behaviours is monitored by tracking the most likely trajectory through such a learned model, while possible failures in execution are identified as deviations from common patterns of trajectories within the learned models. We present results for our experiments with a model learned for a robot behaviour.