A Multi - Task Control Approach for Active-Assisting Rehabilitation

Control of robotic exoskeleton presents an inter-esting research topic in the last years. However, in the field of healthcare, making informed decisions can be challenging because there is not enough data available describing real-world scenarios. In these cases, we can profit from intelligent algorithms that learn from data and aid decision-making to obtain more robust and reliable decisions. In this work, we propose a control approach for active-assisting rehabilitation. It is based on Reinforcement Learning method namely PILCO, short for Probabilistic Inference for Learning Control. It overcomes the high data challenge and does not require an extensive amount of data to design a model that closely approximates the real-world environment. However, PILCO is limited to learn only single task. Therefore, to alleviate the user's workload in the learning process across multiple tasks, we generalize the predicted policy to encompass novel tasks. The novelty in the proposed approach is that the set of tasks used to test the method are predicted from an electromyography signals by the radial basis function (RBF) algorithm. The test of the proposed approach is applied in a simulation environment of a one Degree of Freedom I-DoF elbow-exoskeleton (EE). The simulation results prove the efficiency of the method and show that tasks performed are most often close to target positions.