Online Identification of Pilot Adaptation to Sudden Degradations in Vehicle Stability

Abstract

Time-varying pilot control identification is essential for better understanding of how pilots respond when faced with sudden changes in the dynamics of the vehicle they control, such as when automatic control and stabilization systems disengage or undergo a mode transition. This paper presents the results of a human-in-the-loop experiment performed at TU Delft to test a promising online pilot identification method, based on recursive low-order ARX identification, developed in earlier work. In the experiment, eight skilled participants performed tracking tasks with time-varying vehicle dynamics, where at an unpredictable moment during each tracking run a sudden degradation in vehicle stability was induced. In addition to controlling the time-varying vehicle, participants were asked to indicate when they detected the change in the vehicle dynamics with a button push. This paper compares the effectiveness of two different approaches to detect the moment when pilot adaptation occurs from online identified pilot parameter traces. Overall, the results indicate that the lag in this detection of identified pilot adaptation is equivalent to the subjective detection times, or less. This implies that these online techniques have clear potential for ensuring timely and effective changes in adaptive pilot support systems.