Adaptive Dynamic Programming and Reinforcement Learning Based Navigation of Wheeled Mobile Robots
Term of Award
Master of Science in Applied Engineering (M.S.A.E.)
Document Type and Release Option
Thesis (restricted to Georgia Southern)
Copyright Statement / License for Reuse
This work is licensed under a Creative Commons Attribution 4.0 License.
Department of Mechanical Engineering
Committee Member 1
Committee Member 2
A study is presented on design and implementation of an adaptive dynamic programming and reinforcement learning (ADPRL) based control algorithm for navigation of wheeled mobile robots (WMR). The objectives of the study included modeling of robot dynamics, design of a relevant ADPRL based control algorithm, simulating training and test performances of the controller developed, as well as implementation of the controller using a processor onboard the WMR. First the kinematics and dynamics of a WMR was studied to derive a torque control model based on wheel speeds. The control model was adapted to design an ADPRL based controller for trajectory tracking. The actor and critic modules of ADPRL were based on artificial neural networks (ANN) with inputs as WMR state vector measured at its geometric center. Various training trajectories were used for initial parameter setting of ANN based actor and critic modules with subsequent online learning capability. A Kobuki TurtleBot was used as the WMR platform for both simulation and implementation within the robot operating system (ROS) environment. An open access realistic simulation platform, Gazebo, with ROS integration was used. The ADPRL controller was implemented on the physical WMR through a fast, power-efficient embedded system, Nvidia Jetson TX2, using the open access software platform of TensorFlow for ANN based computation. The effectiveness of the ADPRL controller was illustrated through various test trajectories with comparison of results of test simulations and actual implementation. Both simulation and physical implementation results for the test trajectories met the controller performance criteria.
Popat, Charvi, "Adaptive Dynamic Programming and Reinforcement Learning Based Navigation of Wheeled Mobile Robots" (2018). Electronic Theses and Dissertations. 1811.
Research Data and Supplementary Material