Figure 1

The adaptive spinal-like controller for a two joint planar robot arm. Here, e _cs and e _ce are primary shoulder and elbow motor drives based on weighted visual error from the superior colliculus (SC); e _fs and e _fe are ‘efference copy’ motor predictions based on expected plant behavior from the Renshaw Cells (RC), and K_s and K_e are gains which can internally modify the controller signals. The combined input to the motorneuron (Mn) drives the motors for each segment. Direct intrasegmental connections via interneurons (dashed) are not used in this study since the motors rotate bidirectionally and the dynamic brakes negate the interaction torques. Here, the internal models imbed the combined effect of motor dynamics, sensory-derived mechanical perturbations and internal controller signals - hence predicted states can reflect and compensate for mechanical perturbations. Refer also to Figure  2.