Disturbance Rejection and Roll Over Estimation for Control of Non-Linear Robotic System

Abstract

Accurate state estimation is a foundational requirement for ensuring the safety, stability, and optimal performance of vehicles, whether they are ground vehicles or aerial robotic systems e.g., quadrotors. This research explores the critical realm of state estimation for both ground vehicles and quadrotors, addressing fundamental issues related to vehicle control, safety, and performance enhancement. In the domain of ground vehicles, the precise estimation of the roll angle is paramount for advanced applications, including active anti-roll bars. Traditional methods for attitude estimation have been computationally intensive and reliant on costly techniques like dual antenna global positioning systems (GPS). To tackle this challenge, this research employs a multi-phase approach. In the first phase, 3-DOF vehicle roll dynamics model is deployed along with Luenberger and Sliding Mode Observers to estimate the vehicle's roll angle. The validation is performed against the commercial software CarSim®. The second phase involves the implementation of Complementary and Kalman Filters for roll and pitch angle estimation of ground vehicle, which are independently applied to measure data under different terrains at various frequencies. The dissertation culminates in the proposal of a cost-effective solution to mitigate the risk of vehicle rollovers, emphasizing the practicality and efficiency of the approach through reduction of root mean square error (RMSE) and sample time. Shifting focus to the domain of quadrotors, state and parameter estimation is equally crucial for stable flight, intricate maneuvers, and responsiveness to external disturbances. The fusion of state estimation with advanced control systems, particularly the Sliding Mode control scheme, is explored. Traditional gain tuning for nonlinear systems like quadrotors has been laborious, prompting the integration of Deep Reinforcement Learning (RL) techniques. A comprehensive 6-degree-of-freedom (6-DOF) nonlinear quadrotor model is employed, where aerodynamic coefficients are estimated using the Blade Element Momentum Theory (BEMT). Lyapunov theory and RL optimization are leveraged to ensure system stability, mitigating chattering effects in control inputs. Extensive simulations demonstrate the remarkable effectiveness of this approach, notably reducing the root mean square error (RMSE) during trajectory tracking. The expected results of this research project include the development of innovative and affordable methods for determining roll angle in ground vehicles and attitude in quadrotors, resulting in improved safety and efficiency of vehicle navigation. Additionally, the incorporation of Deep Reinforcement Learning techniques in quadrotor control theory will facilitate the autonomous tuning of Sliding Mode Controllers, resulting in enhanced control performance and flexibility. In general, the results of the research hold great potential for enhancing the current capabilities of vehicle and quadrotor control technologies, ultimately leading to the development of improved and self-governing transportation systems that are safer and more efficient.This research abridges the theory and application, thus providing innovative solutions for real-world challenges in vehicle control and stability.