2020/05/26 by Geoffrey A. Clark, Joseph Campbell, Clark, Geoffrey +7 · 1 citation
Engineering · Health Professions · #Artificial Intelligence (cs.AI) #Balance, Gait, and Falls Prevention #FOS: Computer and information sciences #Machine Learning (cs.LG) #Muscle activation and electromyography studies #Prosthetics and Rehabilitation Robotics #Robotics (cs.RO)
paper · pdf · doi:10.48550/arxiv.2005.13139
openalex publication_date 2020/05/26 · openalex created_date 2022/07/24 · openalex updated_date 2026/07/28
We propose in this paper Periodic Interaction Primitives - a probabilistic\nframework that can be used to learn compact models of periodic behavior. Our\napproach extends existing formulations of Interaction Primitives to periodic\nmovement regimes, i.e., walking. We show that this model is particularly\nwell-suited for learning data-driven, customized models of human walking, which\ncan then be used for generating predictions over future states or for inferring\nlatent, biomechanical variables. We also demonstrate how the same framework can\nbe used to learn controllers for a robotic prosthesis using an imitation\nlearning approach. Results in experiments with human participants indicate that\nPeriodic Interaction Primitives efficiently generate predictions and ankle\nangle control signals for a robotic prosthetic ankle, with MAE of 2.21 degrees\nin 0.0008s per inference. Performance degrades gracefully in the presence of\nnoise or sensor fall outs. Compared to alternatives, this algorithm functions\n20 times faster and performed 4.5 times more accurately on test subjects.\n