Learning Evolutionary Strategy for a Mobile Manipulator in Imitation Learned Tasks

A new algorithm based on Evolutionary Strategies is proposed for finding a robot manipulation path. Next scenario is considered: Given a learned Manipulation Path in the space of configurations, a real-time optimal path is calculated when mobile robot base is in a different position and orientation near to the original localization. The optimization problem is formulated as the minimization of the end-effector position and orientation error so as to ensure convergence towards the learnt path taking into account a time constraint. To solve the optimization problem an algorithm based on the basic Evolution Strategies (ES) scheme is used. ES is a stochastic direct search optimization method based on the evolution of a candidate solution population in an iterative process of mutation and selection. The algorithm avoids singularities since it does not require the use of the Jacobian matrix in the kinematic inversion. The methodology presented is validated in a simulation environment with the mobile manipulator MANFRED, developed in our lab.