In this study we represent the state of art deep neural network models for time series data to be applied in the field of robot learning from human demonstration. We collected dataset for human motion during peg in the hole task using motion capture system alongside with F/T sensor to record human arm joints movements and applied forces/torques respectively. We then will train the DL elected models on this dataset in order to predict a mapped sets of movements for robotic platform (predicted trajectories). The idea is to let these models extract the pattern from this data set to be able to map human movements and applied forces during assembly tasks to robotic kinematics and dynamics.
Али В. (науч. рук. Колюбин С.А.) Deep Neural Networks for robot learning from human demonstration // Сборник тезисов докладов конгресса молодых ученых. Электронное издание. – СПб: Университет ИТМО, [2023]. URL: https://kmu.itmo.ru/digests/article/10226