Learning Human-Robot Collaboration Insights through the Integration of Muscle Activity in Interaction Motion Models

摘要

Recent progress in human-robot collaboration makes fast and fluidinteractions possible, even when human observations are partial and occluded.Methods like Interaction Probabilistic Movement Primitives (ProMP) model humantrajectories through motion capture systems. However, such representation doesnot properly model tasks where similar motions handle different objects. Undercurrent approaches, a robot would not adapt its pose and dynamics for properhandling. We integrate the use of Electromyography (EMG) into the InteractionProMP framework and utilize muscular signals to augment the human observationrepresentation. The contribution of our paper is increased task discernmentwhen trajectories are similar but tools are different and require the robot toadjust its pose for proper handling. Interaction ProMPs are used with anaugmented vector that integrates muscle activity. Augmented time-normalizedtrajectories are used in training to learn correlation parameters and robotmotions are predicted by finding the best weight combination and temporalscaling for a task. Collaborative single task scenarios with similar motionsbut different objects were used and compared. For one experiment only jointangles were recorded, for the other EMG signals were additionally integrated.Task recognition was computed for both tasks. Observation state vectors withaugmented EMG signals were able to completely identify differences acrosstasks, while the baseline method failed every time. Integrating EMG signalsinto collaborative tasks significantly increases the ability of the system torecognize nuances in the tasks that are otherwise imperceptible, up to 74.6% inour studies. Furthermore, the integration of EMG signals for collaboration alsoopens the door to a wide class of human-robot physical interactions based onhaptic communication that has been largely unexploited in the field.