Robot Self-Calibration of Nonlinear Joint Dynamics

Title: Robot Self-Calibration of Nonlinear Joint Dynamics
Author: Axel Tired
Date & Time: June 8th, 11:00-12:00
Location: Seminar Room M 3170-73 in the M-building, LTH
Supervisor: Björn Olofsson, Denis Störkle (Cognibotics)
Examiner:  Yiannis Karayiannidis


Abstract:
The demand for accurate industrial robots has driven an increasing need for accessible and reliable calibration methods. Robot manipulators operating in environments where large external forces are present, or carrying heavy tools, suffer from inaccuracies caused by joint elasticities, both from external loads and from gravity acting on the robot links themselves. Calibration methods for identifying joint stiffnesses are therefore of great importance to compensate for these effects and ultimately obtain more accurate robot models.The clamping method, where a robot arm is attached to a rigid point in the environment and torques are exerted on the joints using the robot's own motors, provides a straightforward approach to joint stiffness identification. However, the assumption of a completely rigid attachment point is not always valid, which may corrupt the identified joint stiffnesses. To address this limitation, a flexible clamping device extends the idea of clamping by serving as both a rigid attachment point and a measuring device. This thesis investigates a method for identifying joint stiffnesses using the flexible clamping device, evaluated through MapleSim simulations. The feasibility of the method is further assessed by applying the identification method to experimental data collected from a physical experiment. Under idealized conditions in simulations, the method correctly recovers the relationship between torque and joint deflections, but does not show consistent accuracy when using a different configuration of the flexclamp, where a lower stiffness results in increased motion at the end-effector. When other joint uncertainties such as passive joint elasticities are included, the identification deteriorates. Furthermore, when translating the method to a physical experiment, additional complexity arises from practical challenges, such as torque limitations in the robot arm. The results suggest that while the method shows promise as an extension of the clamping method, further work is needed to investigate the flexclamp's full ability to identify additional joint properties. In particular, utilizing force measurements from the flexclamp for identification of passive joint elasticities, and more thoroughly investigating how different clamping configurations can aid the identification procedure, are identified as promising directions for future work.