Multi-Objective control through evolutionary neuro-controller for interactive mobile robot manipulator

Effective control of autonomous robots evolving in dynamic and unpredictable environments is a hard problem. The challenging issue is to achieve this control while the robot is performing a constrained and an interactive task. Namely, our aim is endow a mobile manipulator by controllers allowing safe and effective interactions with humans. This multi criteria optimization problem includes the coordination of two or more behaviors, in order to reach global goal. We propose in this paper a new concept to achieve multi-behaviors of catching objects and avoiding moving obstacles simultaneously for autonomous mobile robot arm. The mobile robot is equipped with an arm of 7 DOF. The robot´s learns to avoid obstacles by moving (translation and rotation) its base while its arm tries to catch a given target. Multi-behaviors are controlled by a combination of Continuous Time Recurrent Neural Networks (CTRNN) and Multi-Objectives Genetic Algorithms (MOGA). Furthermore, the Multi-Objective optimization helps to use one genotype for both behaviors to evaluate the CTRNN´s circuit parameters. The results we obtained show that the best individual has a good fitness, meaning that the robot can catch the target with the arm and avoid moving obstacle with the base simultaneously.