Intelligent Robotic Systems – Status of the research at the Laboratory for Robotics and Artificial Intelligence within the Department of Production Engineering

Abstract

In this lecture we present an overview of the recent research efforts in the field of intelligent robotic systems conducted within the Laboratory for Robotics and Artificial Intelligence at the Department of Production Engineering (KaProm), University of Belgrade – Faculty of Mechanical Engineering. The presentation of research results is divided in two main topics. The first topic covers Image-based visual servo (IBVS) control algorithm as one of the approaches used to control the motion of robot manipulators in structured manufacturing environment. Here, we summarize results on the robot motion control in which the information from one camera placed directly onto the robot's end-effector is used within the control loop in order to ensure the desired position of the manipulator. In order to eliminate measurement and modelling errors, a novel intelligent visual servo controller for a robot manipulator using neural network Reinforcement Learning is presented. By implementing machine learning techniques into the vision based control scheme, the robot is enabled to improve its performance online and to adapt to the changing conditions in the environment. Two different temporal difference algorithms (Q-learning and SARSA) coupled with neural networks are developed and tested through different visual control scenarios. Real world experiments are conducted on a 6 DOF NeuroArm Manipulator System and a low-cost camera in an eye-in-hand configuration. The experimental results demonstrate that the proposed method can provide a high accuracy of a manipulator positioning in a situation when the low resolution image is used. The second topic covers intelligent material transport system which is an important part of every modern manufacturing environment. Automated guided vehicles (AGVs) and intelligent mobile robots are a common choice made by many companies for scheduled material transport in manufacturing systems. Having in mind that modern manufacturing environments are exposed to various unpredictable conditions, there is a need that manufacturing environments contain intelligent robotic systems able to learn and to improve their behavior in real time. Therefore, we present a novel method that integrated Learning from Demonstrations (LfD) methodology and chaotic biologically inspired optimization algorithms for reproduction of desired mobile robot motion trajectories. In other words, we suggest nature-inspired learning control approach that enables nonholonomic mobile robot to learn from human teacher and/or robot teacher through demonstrations or observations. In our research work, we have proposed and implemented four different chaotic methods, namely Chaotic Bat Algorithm, Chaotic Firefly Algorithm, Chaotic Accelerated Particle Swarm Optimization, and Chaotic Grey Wolf Optimizer (CGWO) in order to evaluate complex trajectories with different length and unequal number of actuator commands. The optimization objective was to produce such sequence of mobile robot actuator commands that generate minimal error in the final robot pose. The proposed approach is evaluated through the real world experiments carried out on a real mobile robot system (Khepera II mobile robot, low resolution camera, and gripper) in an indoor structured environment. The experimental results show that the mobile robot realized wanted trajectory with minimal error in the final robot pose and successfully finishes the transportation task.