Experimental Identification of the Inverse Dynamic Model: Minimal Encoder Resolution Needed Application to an Industrial Robot Arm and a Haptic Interface

Abstract

To develop appropriate control laws and use fully the capacities of robots, a precise modelization is needed. Classic models such as ARX or ARMAX can be used but in the robotic field the Inverse Dynamic Model (IDM) gives far better results. In this model, the motor torques depend on the acceleration, speed and position of each joint, and of the physical parameters of the link of the robots (inertia, mass gravity, stiffness and friction). The parametric identification estimates the values of these last parameters. These estimations can also help to improve the mechanical conception during retro-engineering steps... It comes that the identification process must be as accurate and reliable as possible. The most popular identification methods are based on the least-squares (LS) regression “because of its simplicity” (Atkeson et al., 1986), (Swevers et al., 1997), (Ha et al., 1989), (Kawasaki & Nishimura 1988), (Khosla & Kanade 1985), (Kozlowski 1998), (Prufer et al., 1994) and (Raucent et al., 1992). In the last two decades, the IRCCyN robotic team has designed an identification process using IDM of robots and LS regressions which will be developed in the second part of this chapter. This technique was applied and improved on several robots and prototypes (see Gautier et al., 1995 – Gautier & Poignet 2002 for example). More recently, this method was also successfully applied on haptic devices (Janot et al., 2007). However, it is very difficult to know how much these methods are dependent on the measurement accuracy, especially when the identification process takes place when the system is controlled by feedback. So, we ignore the necessary resolution they require to produce good quality and reliable results. Some identification techniques seem robust with respect to measurement noises. They are called “robust identification methods”. But even if they give reliable results, they are only applied on linear systems and, overall, they are very time consuming as can be seen in (Hampel, 1971) and (Hubert, 1981). Finally, it seems difficult to apply them on robots and we do not know how much they are robust with respect to these noises.