Bias Estimation and Gravity Compensation for Wrist-Mounted Force/Torque Sensor

Robot force-controlled task executions require an accurate perception of the contact force/torque. When a six-axis force/torque (F/T) sensor is attached to the robot wrist, the compensation precision is affected by many non-contact features, such as the end-effector gravity, inertial, centrifugal, Coriolis forces, and the mechanical errors of the sensor. However, it is complicated to identify all parameters at the same time. Some features will be ignored in practice applications, e.g., surface finishing, human-robot interaction, and haptic applications. In this article, a novel end-effector gravity compensation method for the F/T sensor is proposed, which identifies the end-effector gravity, its center, and the biases (the rotation between the sensor and robot, the robot installation declination, the bias of sensor). Our key is the rotation calibration while the transformation between the F/T sensor and robot is unknown. Simulation and experimental results show that the proposed method works effectively. In a wrist-mounted F/T sensor system, compared to the previous limited method, the compensation error in common force control channel $F_{z}$ by our method diminishes 1.0% of the end-effector gravity (i.e., 21.7% relative improvement).