Combined Kalman Filter-Based External Wrench Estimator with Proprioceptive Sensing for Legged Robot

State estimation is critical for legged robots to explore environments and interact with the external world. However, existing research lacks methods that achieve both high accuracy and real-time performance. To address this, we propose a combined estimation framework based on the generalized momentum Kalman filter (GMKF) and the error-state Kalman filter (ESKF), aiming to accurately and frequently estimate external disturbances during dynamic interactions with the environment. The GMKF is used to estimate ground reaction forces (GRFs), which are then fed into the ESKF to obtain the robot's basic states and external wrenches. The proposed framework is experimentally validated on the BQR3 quadruped robot, demonstrating its effectiveness and superiority.