摘要：This paper describes a novel method for time-varying identification of Human Controller (HC) manual control parameters (called UKF-FPV), based on a steady-state (constant state covariance) Unscented Kalman Filter (UKF). This approach requires noa prioriassumptions on the shape of HC parameter variations, which is a potential advantage over state-of-the-art methods such as the recently proposed MLE-APV approach, for which a sigmoid-shaped parameter variation is assumed. For a scenario where an HC performs a single-loop compensatory tracking task with time-varying controlled system dynamics, both identification methods are compared using Monte Carlo simulations and human-in-the-loop experiment data. Despite some lag in the HC parameter traces of UKF-FPV, the identification results and the HC model quality-of-fit obtained with both methods were found to match well for both the simulation and experiment data. For the experiment data, UKF-FPV even revealed clear "local" changes in HC parameters not captured by the MLE-APV approach, which confirms that HCs adaptunpredictablyeven in what are considered time-invariant conditions. Overall, the results show that an identification method that requires noa prioriassumptions on HC parameter variations is of critical importance for a complete analysis of time-varying HC behaviour.
