Resilient behavior through controller self-diagnosis, adaptation and recovery

We explore robot behavior recovery through a process akin to self-reflection. A robot contains two controllers: A primary "innate" reactive controller, and a secondary "reflective" controller that can observe, model and control the primary controller. The reflective controller adapts the innate controller without access to the innate controller's internal state or architecture. Instead, the reflective controller models the innate controller and then synthesizes input/output filters that adapt the innate controller's existing capabilities to new situations. The innate controller is subjected to a variety of sensory, motor, and internal control damage scenarios. The reflective controller diagnoses the level of failure using a self-model and the observed sensorimotor time-series data and is able to recover performance.