Probabilistic Adaptive Control for Robust Behavior Imitation

In the context of learning from demonstration (LfD), trajectory policy representations such as probabilistic movement primitives (ProMPs) allow for rich modeling demonstrated skills. To reproduce a learned skill with real robot, feedback controller is required to cope perturbations and react dynamic changes in environment. this letter, we propose generalized control approach that merges movements action reproducing behavior. We show our can be easily employed, outperforming both original constant gains. Furthermore, proposed able solve dynamically changing tasks by behavior Gaussian mixtures introducing variables. demonstrate capability experiments simulation teaching 7-axis Franka Emika Panda robot drop ball into moving box only few demonstrations.