Affiliations: [a] Computer Engineering Department, Faculty of Engineering, Yazd University, Yazd, Iran
| [b] Department of Engineering, University of Fasa, Fasa, Fars, Iran
Abstract: This paper proposes a framework for trajectory tracking of wheeled robots in indoor environment. Being robust against uncertainty and scalability to large environments are essential factors for this task. Here, it is supposed that the robot is only equipped with a vision system e.g. a Kinect camera. Generally, some challenges in this problem are: determining suitable control architecture, adjusting the parameters of this architecture according to the given purposes, extracting proper features from high-dimensional input images. In this paper, using deep learning methods the proper features are extracted. The controller is designed based on weighted sum of these features. A new method to combine supervised learning and reinforcement learning is introduced to adjust the proposed controller parameters. The mobile robot and the experimental environments are established on the WEBOTS and MATLAB co-simulation platform. The simulation experimental results indicate that the designed control is robust and effective for tracking trajectories in the indoor environment.
Keywords: Depth data, mobile robot, reinforcement learning, trajectory tracking
