Force feedback design and analysis of central pattern generator model based on echo state networks

A data-driven central paatern generator (CPG) model oriented to robot control was studied in this paper, and the design method about force feedback units was discussed in detail. The robot under the control of such new CPG models would have autonomous capability, and be also adaptive to changing environments. The CPG model was constructed based on echo state networks (ESN) and the control vector was estimated with perturbation methods. The external force was transformed into displacement according to hybrid position/force control principle and then used as feedback errors. The force feedback unit of CPG models was realized with a simple control strategy, for example proportional plus integral olus derivation (PID) control. The feedback unit was internalized into the ESN states by ridge regression estimation. The shape and position of the attractors were altered to equilibrate the external force and save energy consume. The above methods were demonstrated with the open motion capture database.