Situation Assessment for Sensor-Based Recovery Planning
A. Bouguerra, L. Karlsson and A. Saffiotti
We present an approach for recovery from perceptual failures, or more
precisely anchoring failures. Anchoring is the problem of connecting
symbols representing objects to sensor data corresponding to the same
objects. The approach is based on using planning, but our focus is not
on the plan generation per se. We focus on the very important aspect of
situation assessment and how it is carried out for recovering from
anchoring failures. The proposed approach uses background knowledge to
create hypotheses about world states and handles uncertainty in terms of
probabilistic belief states. This work is relevant both from the
perspective of developing the anchoring framework, and as a study in
plan-based recovery from epistemic failures in mobile robots.
Experiments on a mobile robot are shown to validate the applicability of
the proposed approach.