Combined System Identification and Stat...

Precise system identification is an important aspect of adequate control design and parameter definitionto allow for accurate and reliable navigation. While this iswell known in robotics, the community working with smallrotorcraft Unmanned Aerial Vehicles (UAVs) has yet to discoverthe benefits. In contrast to existing work, which often performsoffline or deterministic (i.e. closed-form) system identification,we present a probabilistic approach to the online estimation of system identification parameters and self-calibrationstates. Instead of decoupling system identification and stateestimation for vehicle control, we merge the entire processinto a holistic probabilistic framework to allow self-awarenessand self-healing. Our observability analysis shows that mostof the system identification parameters are observable andconverge quickly to the optimal value using a combination ofinertial cues, dynamic modeling, and an additional exteroceptivesensor. We support our theoretical findings with extensive testssimulating realistic data in Gazebo.