Walking Control Using Adaptive Oscillators Combined with Dynamic Movement Primitives

We address the problem of generating and stabilising a periodic walking pattern for a biped robot. A novel control framework based on adaptive oscillators combined with dynamic movement primitives was proposed. The proposed control system is a multi step process, where in the first step the system learns the dynamic walking trajectory which is demonstrated by imitating a spring loaded inverted pendulum (SLIP) controller. The learning process is completely autonomous and it is done in real time. In the second step the periodic dynamic pattern is executed using a low gain feedback controller combined with feed forward torque/force control signals. This allows compliant behaviour, smooth interaction with unstructured environment and generates stable ”open-loop” walking pattern. Compared with the SLIP walking control the proposed approach is more robust since it can produce stable walking pattern on a wider interval of forward walking velocity. The proposed control also allows smooth modulation of forward walking velocity