In this work, experimental data is used toestimate the free parameters of dynamical systemsintended to model motion profiles for a robotic system.The corresponding regression problem is formedas a constrained non-linear least squares problem.In our method, motions are generated via embeddedoptimization by combining dynamical movementprimitives in a locally optimal way at each time step.Based on this concept, we introduce a model predictivecontrol scheme which allows generalization overmultiple encoded behaviors depending on the currentposition in the state space, while leveraging the abilityto explicitly account for state constraints to the fulfillmentof additional tasks such as obstacle avoidance.We present a numerical evaluation of our approachand a preliminary verification by generating graspingmotions for the anthropomorphic Shadow Robothand/arm platform.
Funding Agencies:
HANDLE - European Community ICT-231640
ROBLOG - European Community ICT-270350