Interaction Force and Torque Detection for Robotic Manipulators Based on Kalman Filter With Automatic Calibration and Adaptive Moving Average With Varying Gain

This article proposes an observer capable of estimating the interaction force and torque for single-arm or dual-arm robots without requiring physical force/torque sensors. To reduce the predicted state noise and improve confidence in the calculated joint driving torques, an adaptive moving average (AMA) is developed. The AMA integrates a Hull Moving Average with a variable gain and a weighted moving average with a fixed gain, automatically switching between the two to achieve an optimal balance between estimation accuracy and response time. Furthermore, tuning routines for the AMA and the Kalman filter are introduced, considering measurement noise and dynamic uncertainties. Larger diagonal elements of the covariance matrix of the predicted state noise can be employed and thus the response speed of the observer is improved despite the inherent time lag of the moving averages. The proposed observer was validated on both the single-arm and the dual-arm robotic systems under various scenarios, demonstrating improved estimation accuracy and faster response speed compared to the existing observer. The proposed observer can be applied to robots with serial robotic arms, offering a cost-effective solution for force and torque estimation.