טכניון מכון טכנולוגי לישראל
הטכניון מכון טכנולוגי לישראל - בית הספר ללימודי מוסמכים  
M.Sc Thesis
M.Sc StudentSlobodinsky Alexander
SubjectComparison of Adaptive Control Methods for Robots
DepartmentDepartment of Mechanical Engineering
Supervisors Professor Emeritus Yehoshua Dayan
Professor Moshe Shoham
Dr. Strasberg Shai


Abstract

Most of recent publications dealing with robot adaptive control algorithms present the algorithm, prove the stability of the algorithm and demonstrate it by simulation of a single robotic arm or up to 2-3 DOFs. Few attempts to compare the different methods were presented since 1986. The present thesis is aimed at carrying further and extending the preceded works, and comparing between the different adaptive control methods for robot manipulators. Sixteen different methods were chosen for the comparison. The algorithms were characterized by wide variety of different properties, such as centralized vs. decentralized control, linear vs. non-linear models, different adaptation mechanism, etc. The different algorithms were applied, by simulation, to a 5 DOF industrial robot, Mitsubishi RV-M2, along two trajectories at different velocities and different conditions (no disturbance operation, friction disturbances, artificial disturbances, sudden load changes, etc.). Two different cost functions were defined and weight assigned to errors, moments, computation cost and stability. The cost functions are calculated for the different control algorithms considered and serve as the basis for selection of optimal ranges of the systems parameters. An attempt to select the optimal control parameters has been made for each of the algorithms, according to the desired trajectory. Several criteria were selected for the examination and comparison. These are: stability, level of performance, robustness to changes and uncertainties and the cost of applying each algorithm. Weights were selected according to the quantitative criteria, which together with the qualitative criteria form the basis for weighing and selecting the best algorithm for a particular task and a specific robot. The best are the following three algorithms: Sadegh and Horowitz 1990, Zergeroglu et al. 2000, Santibanez and Kelly 1999.