Sensorless External Force Estimation for Robotic Arms via a Confidence-Aware Soft-Switching Hybrid Filter

Abstract

Accurate and robust external force perception is a fundamental prerequisite for safe physical human-robot interaction in complex dynamic environments. While sensorless force estimation utilizing generalized momentum circumvents the spatial and economic constraints of dedicated hardware, existing single-model filtering approaches struggle to balance steady-state fidelity with rapid transient response, often suffering from convergence lag or severe overshoot during abrupt impacts. In this paper, a Confidence-aware Soft-switching Hybrid Filtering approach (CSHF) is proposed to address the trade-off between estimation precision and robustness. Leveraging the generalized momentum observer (GMO) as the foundational residual generator, a dynamic collision confidence evaluation mechanism based on a non-linear activation function is introduced. By means of this mechanism, an adaptive, smooth transition is enabled between the Kalman Filter (KF) for high-fidelity steady-state smoothing and the Particle Filter (PF) for non-linear, transient impact tracking. Extensive simulations conducted on a 7-DOF xArm7 manipulator demonstrate that the proposed method significantly reduces both root-mean-square errors (RMSE) and mean-absolute errors (MAE) across diverse excitation scenarios. The limitations of traditional Kalman variants are effectively overcome by the CSHF approach, providing a reliable and low-cost force perception solution to enhance compliant robot control.