Analysis of the role of proprioceptive information during arm movements using a model of the human arm.

Using a model for the neuromuscular control of human arm movements, the possible roles of different proprioceptive signals are analyzed. The control model is represented by a neural network and includes both feedback and feedforward control modes. After a learning process, the controller regulates a wide range of arm movements. Evaluation of the roles of different afferent signals shows that sensed muscle forces are important to achieve accurate control of fast movements. For a moderately high look delay (50 ms), velocity feedback is not essential, but for small look delays (0 to 25 ms) an increased performance is attained by feedback of velocity. Position sense is essential to prevent steady-state errors. The arm impedance is affected considerably by the delay in the control loop and by the configuration of the motor control system. The achieved relation between muscle length and force is similar to the invariant characteristics laying at the basis of the equilibrium-point (EP) hypothesis. However, control of fast movements on the basis of EP along is not feasible, but required feedforward control. During training in a velocity-dependent force field, the impedance of the arm increases at first, due to enhanced cocontraction. Subsequently, both impedance and movement errors decrease, indicating a successful representation of the changed inverse dynamics.