Optimal Control of an n-Legged Robot

Abstract

This paper presents a method for controlling the dynamic balance of legged robots using optimal state feedback. Rather than being restricted to a specific number of legs, the method considers the general case of a machine with n legs. The analysis starts with a non-linear dynamic model of a general robot and a set of equations representing the constraints on motion imposed by those feet in contact with the ground. These equations are used to derive a state-space model of order proportional to the number of degrees of freedom of the system, which will vary with the current constraint conditions. An optimal feedback gain matrix is then calculated for the linear model using standard techniques. The choice of operating point and optimization parameters is discussed. The effectiveness of the method is illustrated through simulation responses obtained .for a biped model under both single and double support constraint conditions.

Research at the MEC

“…Generic knowledge arising from this research area includes novel types of recurrent and self-organising neural networks, genetic algorithms for small chromosome populations and fuzzy logic backward reasoning methods. [17][18][19][20][21][22][23][24][25][26][27][28][29][30][31][32][33] References [21] and [33] are monographs that both had to be reprinted immediately after first publication due to high demand. Reference [21] on neural control is now in its fourth printing without needing revision.…”

A proposed hybrid neural network for position control of a walking robot

“…State feedback laws have been suggested [5], since they are a natural control method for multi-variable systems. A method of deriving the Linear Quadratic Regulator (LQR) for an n-legged robot that could be adapted for bipeds has been presented by Channon et al [6]. A neural network which is a recurrent hybrid network has been employed to control a bipedal robot.…”