A natural approach to efficient motor learning in goal-directed motion tasks

Introduction: Parkinson's, stroke, and other neurological diseases may cause significant problems in human motion behaviour. In particular, such diseases affect the control functions in goal-directed, voluntary movements that normally are performed optimally as regards motion speed, positioning accuracy, and energy expenditure. The control functions (neural signals to muscles) are to be re-learnt and re-optimised with respect to these performance indices. In our study, a natural approach for efficient motor learning in goal-directed motion tasks is proposed. It is based on novel concepts and underlying principles of robot dynamics, optimal control theory, and biological cybernetics.

Method: Optimal control functions have a triphasic shape and a set of key parameters is found to be necessary and sufficient for describing them. Those are intrinsic parameters human has to learn in dynamic point-to-point motion tasks, [1]. The control learning scheme has the following main steps, [2]: 1) parameterise test control functions; 2) select most appropriate pairs of control parameters and controlled outputs; 3) make corrections in the control parameters until reach the target, applying an optimal, convergent and natural learning algorithm.

Results: Using realistic mathematical models, our motor learning approach was applied to motion tasks like reaching movements, Fig. 1, and performing steps, Fig. 2. In the latter case, we decomposed the task to perform a step into two point-to-point leg movements. In the computer simulation, we verified that the learning control parameters converge and the number of trials is very small. In addition, we did some real (able-bodied) experiments with rapid aiming movements of the arm and they confirm the feasibility and efficacy of the proposed approach. Thus we have now good evidence in proving our novel concepts for optimal control learning and that is very important for designing reliable control strategies in neurorehabilitation.

Discussion: The neural structures that compute the required control forces are the so-called internal models presenting a fundamental part of the voluntary motor control. We believe that the proposed approach can be used to rebuild such models (cortical reorganization) by proper training procedures. Our approach can also be used for the purposes of neuro-muscular rehabilitation, with assistive robotic devices applied or in functional electrical stimulation. In the latter case, optimising the timing sequence for stimulating muscles may produce smoother and more accurate movements. Another interesting problem is the design of brain-computer interfaces for direct control of human/robot motion and this is a subject of ongoing research.

Abb. 1

Abb. 2

References:

[1]. Karniel, A. and G. Inbar, 1997. A model for learning human reaching movements. Biological Cybernetics, Vol. 77(3), pp. 173 – 183.

[2]. K. Kiryazov and P. Kiriazov, Efficient learning approach for optimal control of human and robot motion, in: Emerging Trends in Mobile Robotics, Eds: H. Fujimoto, M O Tokhi, H. Mochiyama G S Virk, World Scientific Publishing Co., 2010, pp 1219 – 1227