Improving Learning Performance in a Reaching Task by Real-time Adaptation of Augmented Error Feedback Rules

Abstract

Neurologically impaired patients can regain motor function by engaging in rehabilita- tion. Currently there is no conclusive evidence that robotic rehabilitation has better clinical results than conventional rehabilitation but robotic rehabil- itation has the potential to increase efficiency and patient motivation, justifying improvement of reha- bilitation robotics. A variety of approaches to de- sign the interaction between a robotic trainer and the patient is used. For the purpose of improving rehabilitation robotics, a new adaptive algorithm is proposed. Assistive algorithms seem suitable for training impaired patients but prove difficult to val- idate with healthy subjects. For healthy subjects, error augmentation is shown to be more effective for learning. In an experiment, 13 healthy subjects performed a reaching task while strapped to an up- per extremity exoskeleton. During this task they were subject to an adaptive augmented error feed- back controller. Subjects were divided into three groups in which one of each, or both the follow- ing parameters of a force field were adapted: dead band width and divergent force field strength. Per- formance was measured as amount of deviation from a straight line between two targets. Adapt- ing dead band width results in a better movement performance than adapting force field strength (p = 0.0069). Adapting force field strength in addition to adapting dead band width did not improve move- ment performance (p = 0.9960). It is concluded that an adaptive augmented error feedback mechanism can improve movement performance in a reaching task with healthy subjects.