Joint torque feedback is useful in serial manipulator control algorithms for contact control, collision detection, performance analysis, etc. For example, the predicted torque can be compared to the measured torque so the system can respond to unexpected or unmodeled physical inputs. The input current to the joint motors can be used to estimate the input torque if the motor parameters are well-understood. However, in a closed commercial system, the motor parameters are often proprietary or unknown. Also, systems that sense or estimate motor torques instead of the joint torques require compensation for gear train losses.
In this work, we propose a method for mapping the measured motor current to the joint torque on a serial manipulator without joint torque sensors, thus advancing the potential to implement torque feedback algorithms such as collision detection on any industrial robot with joint position and motor current feedback. This new torque estimating technique (as opposed to using Newton-Euler dynamics) allows for sensing of external forces in collision detection applications for a position controlled robot.
The method requires knowledge of the robot link centers of mass, masses, and inertias and that the motor currents and joint positions can be measured. The joint torques due to gravity, inertia, and Coriolis are estimated by the Newton-Euler method using the system geometry, link masses, and the measured joint positions. A method for estimating friction losses using only the current and the predicted joint torque is demonstrated. The measured current, less estimated friction, is then mapped to the joint torque.
The validity of the black box joint torque estimating model was demonstrated using two Motoman SIA-5D manipulators with a 3rd party controller provided by Agile Planet. The joints of the robot were moved through a variety of test motions with known joint torque characteristics (as calculated using Newton-Euler dynamics). Estimated joint torques are similar to the calculated torque. Physical significance of the torque is validated by comparing the estimated torque to the calculated torque generated by a known force. The feasibility of the estimated torque error to force detection is discussed in terms of improving the safety and deployment options for industrial robotic systems.