Automatic identification of arm-manipulator mechanics
Abstract
Accurate robotic arm operation typically relies on detailed models, often derived using the Denavit-Hartenberg (DH) convention. However, DH requires precise physical descriptions. This study explores using machine learning to estimate a robot's physical parameters solely from joint angle time-series data, eliminating the need for pre-defined models. This approach enables large-scale system identification and presents an avenue for advancing ML in robotics. The developed algorithm was successfully validated on both simulated and physical robotic arms, demonstrating its potential for practical application.
Primary Faculty Mentor Name
Nick Cheney
Status
Undergraduate
Student College
College of Engineering and Mathematical Sciences
Program/Major
Electrical Engineering
Primary Research Category
Engineering and Math Science
Automatic identification of arm-manipulator mechanics
Accurate robotic arm operation typically relies on detailed models, often derived using the Denavit-Hartenberg (DH) convention. However, DH requires precise physical descriptions. This study explores using machine learning to estimate a robot's physical parameters solely from joint angle time-series data, eliminating the need for pre-defined models. This approach enables large-scale system identification and presents an avenue for advancing ML in robotics. The developed algorithm was successfully validated on both simulated and physical robotic arms, demonstrating its potential for practical application.
