Sensory-Motor Mapping With A Sequential Network

Abstract

This paper deals with the problem of representing and generating unconstrained aiming movements of a limb by means of a sequential neural network. Targets are specified to the network as sensory stimuli; for each stimulus the network produces a time trajectory of a redundant limb from a starting posture towards the corresponding target. The network was trained using a bell-shaped velocity profile on the trajectories, which is a recurring feature of movements performed by biological systems. We performed a number of experiments, both during and after learning. Our results show that: (i) the task can be learned by a three-layer sequential network; (ii) the network successfully generalizes in trajectory space and adjusts the velocity profiles properly; (iii) the same task cannot be learned by a linear network; (iv) the model is robust to noise on the input signals; (v) the network exhibits attractor-dynamics properties; (vi) the network is able to solve the motor-equivalence problem. A key feature of this work is the fact that the neural network was coupled to a mechanical model of a limb in which muscles are represented as springs. This representation made it possible to deal with the redundancy of the motor apparatus.