Journal of Mechanical Engineering Research Vol. 4(3), pp. 89-99, March 2012 Available online at http://www.academicjournals.org/JMER DOI: 10.5897/JMER11.061 ISSN 2141-2383 ©2012 Academic Journals Full Length Research Paper Robust least square estimation of the CRS A465 robot arm’s dynamic model parameters Azeddien Kinsheel 1 *, Zahari Taha 2 , Abdelhakim Deboucha 1 and Tuan Mohd Yusoff Shah Tuan Ya 1 1 Faculty of Engineering, University of Malaya, 50603, Kuala Lumpur, Malaysia. 2 Faculty of Manufacturing Engineering and Management Technology, University Malaysia Pahang, 26300 Gambang, Pahang, Malaysia. Accepted 19 November, 2011 This paper presents the experimental estimation of the barycentric parameters of the CRS A465 robot arm. Three methods are used to estimate the barycentric parameters of the arm; ordinary least square method, weighted least squares method and iteratively reweighted least squares method. The estimation is carried out on the complete robot model and the simplified model where the effect of the product of inertia is ignored. The joints friction is represented by the standard friction model. The obtained results show how the identification methods and the model simplification affect the parameters estimation and joint torque prediction in real system identification. Key words: Identification, least squares, robot, dynamics, barycentric parameters, robust least squares. INTRODUCTION Advanced applications of robotic systems such as robot- assisted surgery require an accurate model of the robotic system for advanced control system design, preoperative planning, process supervision, simulations, and training. Robot models, in turn, require sufficiently accurate knowledge of the parameters of the robot dynamics. Typically, robot technical manuals do not provide all the required data such as link masses, inertia and joint friction. Instead, experimental identification of the dynamic parameters is used. In general, a standard robot identification procedure consists of robot modeling, trajectory generation, data acquisition, signal processing, parameter estimation and model validation (Paul et al., 1983; Raucent et al., 1991; Swevers et al., 1996; Reyes et al.,1997; Kozlowski, 1998; Verdonck et al., 2001; Khalil et al., 2002; Swevers et al., 2002; Grotjahn et al., 2004; Waiboer et al., 2005; Francesc et al., 2006; Radkhah et al., 2007; Khalil et al., 2007; Karahan et al., 2008; Swevers et al., 2007). The dynamic robot model describes the rigid body dynamics by relating the robot motion to the joint torques. For serial robots, the *Corresponding author. E-mail: eng.hakim25@gmail.com. Lagrange-Euler (L-E) and Newton-Euler (N-E) formulations are the methods that have often been used to obtain the inverse dynamic model. For experimental identification, the inverse dynamic model is usually obtained in a canonical form that is linear in the robot inertial parameters. The canonical form allows dynamic parameter to be identified separately from the known geometric and kinematic parameters. The dynamic parameters are grouped into what are known as base parameters (Khalil et al., 2007) or barycentric parameters as defined by Fisette et al. (1996). The categorization and the grouping of the base parameters is done symbolically or by applying a set of rules (Kozlowski, 1998; Khalil et al., 2007). The motion trajectory used to collect motion data is designed, so that all the identifiable parameters are persistently excited. The trajectory optimization problem is addressed in several works (Khosla, 1998; Armstrong, 1989, Reyes et al., 1997). An efficient periodic band-limited excitation trajectory which is easy to process and to provide an accurate estimate is presented by Swevers et al. (1996). To improve the signal to noise ratio, the data measured during robot motion are filtered - preferably with the same filter (Khalil et al., 2002). The arrangement of the robot’s inverse dynamic model in the canonical form allows the use of common