Estimation of the Absolute Orientation of a Five-link Walking Robot with Passive Feet

One of the main objectives of current research on walking robots is to make their gaits more fluid by introducing imbalance phases. For example, for walking robots without actuated ankles, which are under actuated in single support, dynamically stable walking gaits have been designed with success (Aoustin & Formal’sky 2003; Chevallereau et al. 2003; Zonfrilli et al. 2002; Aoustin et al. 2006; Pratt et al. 2001; Westervelt et al. 2003). Both the design of reference motions and trajectories and the control of the mechanism along these trajectories usually require the knowledge of the whole state of the robot. This state contains internal variables which are easy to measure using encoders for example, and also the absolute orientation of the robot with respect to the horizontal plane. For robots with unilateral constraints with the ground, it may not be possible to derive the latter information from internal measurements, as in the case of the absolute orientation of a biped in single support. Of course, technological solutions exist such as accelerometers, gyrometers, inertial units... But the implementation of these solutions is expensive and difficult. In order to overcome these difficulties, we propose to use a state observer which, based on the measurements of the joint variables and on a dynamic model of the robot, provides estimates of the absolute orientation of the walking robot during imbalance phases. This strategy was first validated in simulation for a three link biped without feet, using nonlinear high gain observers and a nonlinear observer based on sliding modes with a finite time convergence (Lebastard et al. 2006a) and (Lebastard et al. 2006b), for walking gaits composed of single support phases and impacts. The main drawback with this family of observers is that, when only some of the degrees of freedom are measured, a state coordinates transformation is necessary to design their canonical form (Gauthier & Bornard 1981; Krener & Respondek 1985; Bornard & Hammouri 1991; Plestan & Glumineau 1997). In this chapter, the observer is an extended Kalman filter and it is applied to SemiQuad, a prototype walking robot built at our institute. SemiQuad is a five link mechanism with a platform and two double-link legs. It is designed to study quadruped gaits where both front legs (resp. rear legs) have identical movements. Its unique front leg (resp. rear leg) acts as the two front legs (resp. rear legs) of the quadruped, so that SemiQuad can be considered as an implementation of a virtual quadruped, hence its