Statistical Analysis Methods and Applications of Errors in Robot Motion Control Experiments

Robot motion control is crucial for high-precision tasks in fields such as industrial manufacturing and surgical operations. However, multi-source errors related to machinery, sensors, environment, and modeling significantly reduce its precision. This paper systematically reviews the statistical analysis methods for such errors in motion control experiments, with a focus on introducing the mathematical modeling and handling strategies of errors. For modeling, it analyzes probability-based error propagation models such as covariance analysis-least squares, Monte Carlo simulation, Taylor series expansion, and non-Gaussian modeling to quantify the transmission of uncertainties in the kinematic chain, as well as spatiotemporal correlation models such as Markov chain integrated stochastic frameworks and multi-source error Bayesian networks to capture the error dynamics coupled with time and space. For processing, it explores three technical directions: first, real-time filtering and state estimation, which includes statistical process control and Bayesian network fusion; second, parameter identification and system calibration, including genetic particle swarm optimization-neural network and Bayesian optimization-random forest; third, robust control and adaptive strategies, including deep learning, dynamic compensation, and federated learning, among others. It compares the applicability of methods. For example, the Monte Carlo method is used for offline nonlinear analysis but has a large computational load; federated learning is used for rapid multi-robot convergence but has high bandwidth requirements to guide selection, and looks forward to future research directions, such as improving robustness in extreme environments.