Adaptive compensation for time-varying uncertainties in model-based control of lower-limb exoskeleton systems

Dear editor, Exoskeletons are widely used in human locomotion assis-tance and strength augmentation to enhance strength and endurance,wherein strategies based on model-based control are extremely sensitive to errors in the dynamic models.Ad-ditionally,the unpredictable human-exoskeleton interaction force can unbalance the pilot.Therefore,learning meth-ods were utilized to improve the accuracy of the dynamic model[1,2].The modeling of dynamic model was decou-pled based on the direction of interest[3].Nevertheless,the time-varying uncertainty of the dynamic model[4]forced the learning algorithm to learn this model at all times.In this study,a novel strategy called adaptive compensation learning (ACL) is proposed to consider the extra interaction forces caused by the time-varying uncertainty,in which the model-based controller is used to amplify the forces caused by the pilot[5],and the compensation strategy generator adaptively generates a compensation strategy for the extra interaction forces.Reinforcement learning can determine the optimal coefficients of the ACL strategy for different pilots and walking patterns.In the learning process,we consider the interaction forces and their changing velocities.The efficiency of the ACL is verified through the comparison experiments,and the experimental results demonstrate that the ACL can achieve better performance.