Uncertainty-aware Safe Adaptable Motion Planning Of Lower-limb Exoskeletons Using Random Forest Regression

Human safety and data security are two of the main concerns that have limited the utilization of deep learning-based techniques in medical robotic applications. Such concerns are amplified by uncertainty in the deep learning run-time predictions. In this paper, we propose a novel framework for incorporating uncertainty analysis that is fast enough to be used in the control loop of a medical robot and that considers both the training and testing phases of the deep learning algorithm. As a case study focusing on the use of a lower-limb exoskeleton to assist the walking of people with disability, we learn the passive human-exoskeleton system’s dynamics using Random Forest Regression (RFR) and quantify the uncertainty level of its prediction. Whereas prior art fed the estimated human-robot interaction torque values to the adaptive Central Pattern Generators (CPGs) to refine the gait trajectories, our contribution is to leverage the knowledge of the predictions’ uncertainty levels to ensure safety in human-robot interaction. Our proposed framework for uncertainty-aware control of medical robots finds the similarities of labels and predictions in the training set using Kullback-Leibler (KL) divergence between input and training distributions and detects out-of-distribution (OOD) data using Mahalanobis distance between test feature and training set. We have tested the proposed method on ExoH3 (Technaid) lower-limb exoskeleton. The experiments were conducted to evaluate the performance of the uncertainty analysis technique. The results demonstrate that our proposed uncertainty analysis technique can detect OOD features resulting in unsafe motion planning. We also showcase the importance and effectiveness of using uncertainty analysis in the lower-limb exoskeleton case study.