New Approach of Neural Network for Controlling Locomotion and Reflex of Humanoid Robot

Abstract

In recent years, several humanoid robotic platforms have been developed, and most of them were dealing with hardware problems to closely replicate the appearance and the motion ability of human beings. Along with this, motion control also has been advanced, and much work has been focused on the dynamics of the robot using the zero moment point (ZMP) approach (e.g. Vukobratovic & Juricic, 1969; Miura & Shimoyama, 1984; Kajita & Matsumoto, 2001; Huang et al., 2001). Huang et al. proposed a method for planning a walking pattern, where the reference trajectory is designed offline for given constraints on the foot and ground, and satisfying a particular ZMP constraint using third order spline functions. More recently, biologically inspired control strategies have been proposed to generate autonomously adaptable rhythmic movement. These are based on a neuronal network, termed a central pattern generator (CPG) (e.g., Griller, 1985; Taga et al., 1991; Taga, 1995) that is capable of generating a rhythmic pattern of motor activity in the absence of sensory input signals. Taga et al. demonstrated that bipedal locomotion can be realized as a global limit cycle generated through entrainment between a neural network consisting of a neural oscillator and the physical system. On the other hand, toward a safe interaction of the humanoid robot with the environment, Morisawa et al. presented a method to generate an emergency stop motion based on the evaluation on the ZMP and the center of gravity (COG). Okada et al. presented a motion emergency system based on attractor design of the nonlinear dynamical system. Huang et al. proposed a feedback sensory reflex, which consists of ZMP reflex, landing-phase reflex, and a body-posture reflex. However, in the presence of large disturbances, the ZMP will have an arbitrary location that can be out of the stable region in spite of the stability of the robot’s upper body. In this chapter, we deal with both the motion patterns generation and reflexes against sudden events. Instead of using complex modeling and dealing with highly non-linear equations (Okada et al., 2005; Huang & Nakamura, 2005), our control strategy in generating rhythmic motion is simply based on piecewise linear oscillators and a small number of easily tunable parameters. The method needs not to satisfy constraints on robot’s foot or ZMP stability margin as in (Huang et al., 2001). It simply uses piecewise-linear functions