Posture control of a cockroach-like robot

Abstract

As legged robots become more animal-like, it is likely that these robots will have many complex limbs with redundant degrees of freedom (DOF). This is especially true when we desire them to move like an animal. Animals are capable of spontaneous and non-stereotyped locomotion, such as turning, swaying, twisting, deliberately falling, jumping, climbing, and running, Therefore, it becomes difficult to provide joint space trajectories, in real-time, for these complex movements when many limbs are simultaneously involved, and when some or all of these limbs contain redundant DOF. When locomotion takes place rapidly, it has been suggested that there is a feedforward control component that involves a proactive, higher level computation in the nervous system. We suggest and demonstrate an intuitive and computationally simple algorithm for controlling the posture of a complex, multileg robot with many redundant DOF. The algorithm avoids inverse kinematics by issuing feedforward force commands to both maintain static posture and generate body motion. In so doing, it is also shown that the multileg mechanics of postural control can be reduced to a simple center-of-pressure representation, or equivalently, an instantaneous virtual leg model.