Autonomous Causally-Driven Explanation Of Actions

We propose a cause-effect reasoning mechanism with which an autonomous system can justify planned actions to a human end user. The mechanism is based on a structure we call a "causal plan graph," which encodes the causal relationships between the actions, intentions, and goals of the autonomous system. Causal chains within this graph can potentially serve as intuitive, human-friendly justifications for the autonomous system's planned actions. A prototype of this mechanism is tested in simulation on a set of planning problems from an autonomous maintenance scenario. We demonstrate empirically that shortest path algorithms can effectively reduce a very large number of possible causal chains to a small, intelligible subset that might reasonably be inspected and ranked by a human. Consequently this work can serve as the basis for an experimental platform for future end user studies with human participants.