Online omnidirectional walking patterns generation for biped robot

This paper proposes a method to plan omnidirectional walking patterns at first. Only the motions of hip and feet are planned. And the motions of other joints can be obtained through inverse kinematics constraints. Then dynamic online walking path and gaits generation are discussed. ANN is used to optimism the generated walking gaits. The ANN is trained offline in the simulation system. Then biped robot can walk to a variational target position without any stopping to turn its body. It makes the movement of robot faster, agiler and more stable. Our proposed approach has been tested in RoboCup soccer 3D simulation system. It is proved to have advantages when acting in dynamic environments.