A recurrent neural network model for generation of humanlike reaching movements

We constructed a recurrent neural network model that can generate human reaching motion. Given a target position, the neural network model is capable of producing muscular activation signals that move the hand to the target endpoint, while solving the problem of musculoskeletal redundancy by dynamic relaxation of the energy functions embedded in the network. The proposed neural network model was integrated with a two-dimensional three-link eight muscle musculoskeletal model of the human arm to simulate arm reaching movements in the horizontal and sagittal planes. Our results demonstrate that the model is capable of generating natural arm movements that have spatiotemporal features, such as a slightly curved hand path and the characteristic bell-shaped velocity profile, that are similar to those of actual human movements. Some aspects of the proposed computational framework might be utilized in the central nervous system for generation of reaching movements.