Session: 07-10-02: Mobile Robots and Unmanned Ground Vehicles

Paper Number: 114223

114223 - Comparison Between Two Distinct Dynamic Modeling Techniques and Three Linear and One Nonlinear Controller for a Universal Omni-Wheeled Mobile Robot: Application Offline Reinforcement Learning Based Navigation 

Mobile robots are a combination of hardware and software components to move in free space. Some of the examples for mobile robots are humanoid robots, unman rovers, entertainment pets and drones. Several fields of research in robotics are emerging, such as wheeled mobile robots, legged robots, flying robots, robot vision, and machine learning in robotics. Mobile robot applications include surveillance, planetary exploration, emergency rescue operations, reconnaissance, petrochemical applications, industrial automation, construction, entertainment, guide, intervention in extreme environments, etc.

Locomotion, perception, cognition, and navigation are fields of research in mobile robots. Locomotion problems are solved by studying the applicable mechanisms of kinematics, dynamics, and control. Robot locomotion not only depends on the mechanisms, but also its environment maneuverability, and controllability. Mobile robots can be categorized depending on the locomotion such as walking, rolling, jumping, running, sliding, skating, swimming, and flying. Wheeled Mobile Robots (WMR) are one of the major categories of land based mobile robots.

Omni-directional locomotion provides WMR additional maneuverability and subsequent productivity. As presented and discussed in Amarasiri et. al. [1], Universal Omni-Wheels are one of the best categories of wheels that can be used to develop a WMR as they can carry more weight, have better maneuverability, and have less complexity of design.

One of the main objectives of the work herein is to use a dynamic model of the mobile robot in a Reinforcement Learning (RL) based navigation algorithm to train an agent offline. For a Reinforcement Learning agent to perform a task it requires copious amounts of learning episodes. Therefore, training is very time consuming. To make training stage faster, it is very important to have a very fast dynamic model and/or controller. In this paper, we compare a traditional Kane’s equations model to a nonholonomic momentum model [2]. Also, we implemented four controllers: Proportional Integral Derivative (PID), Linear Quadratic Regulator (LQR) with Integral action, pole placement, and a full nonlinear Sliding Mode Controller (SMC).

This work clearly shows the advantages and the disadvantages for each of the modeling techniques, and control laws implemented. Ultimately the fastest combination will be used for RL training for path generation in unstructured environments.

Index Terms—Universal Omni-wheels, Kinematic, Inverse Kinematic, Multibody Dynamics, Trajectory tracking, PID controller, Mobile Robot, non-holonomic, Kane’s Equations, Momentum Equations, Sliding Mode Controller, LQR Controller, Pole Placement

[1].  Amarasiri, N., Barhorst, A.A. and Gottumukkala, R., 2022. Robust Dynamic Modeling and Trajectory Tracking Controller of a Universal Omni-Wheeled Mobile Robot. ASME Letters in Dynamic Systems and Control, 2(4), p.040902.

[2].  Barhorst, A.A., 2019. Generalized momenta in constrained non-holonomic systems—Another perspective on the canonical equations of motion. International Journal of Non-Linear Mechanics, 113, pp.128-145.

Presenting Author: Nalaka Amarasiri University of Louisiana at Lafayette

Presenting Author Biography: Nalaka Amarasiri is a Ph.D. student who is working in the area of robot dynamics, controls, and Reinforcement Learning with Dr. Alan A Barhorst and Dr. Raju Gottumukkala. Passionate about working with classical robot controllers alongside Machine learning to make better performance and controllability.

Authors:

Nalaka Amarasiri University of Louisiana at Lafayette
Alan A. Barhorst University of Louisiana at Lafayette
Raju Gottumukkala University of Louisiana at Lafayette