On improving the adaptability of controllers and estimators for mobile robots in challenging operational conditions

Abstract

To achieve full robot autonomy in real-world scenarios, the estimators and controllers designed for robotic systems must encompass adaptability to combat the uncertainties that often arise. Although most existing methods provide sufficient robustness to reject disturbances, systems operating in adverse deployment environments or complex configurations typically require a more sophisticated design to maintain their integrity. Therefore, within the context of mobile robots, this thesis investigates and aims to improve the adaptability of systems subject to these challenging operational conditions. In the first half of the thesis, we delve into the controller design of the Unmanned Underwater Vehicle (UUV), where we propose a novel sensor-based Model Predictive Control (MPC) that accounts for external disturbances and measurement noise. To enhance the applicability of the Disturbance Observer (DO), we first reformulate the Extended State Observer (ESO) with an Inertial Measurement Units (IMU) sensor model, constructing an Error-State Extended State Observer (ESESO). By embedding the ESESO into the prediction model of the nonlinear MPC, we are able to improve the system's trajectory tracking performance. In the second half of the thesis, we examine the problem of Relative State Estimation (RSE) with a heterogeneous Unmanned Aerial Vehicle (UAV)-Unmanned Ground Vehicle (UGV) system. Specifically, the UAV-UGV team is constrained to operate as a Non-Inertial Control Systems (NICS), where the UAV can only be estimated and controlled by the ground-based UGV. This configuration motivates an inquiry into the Unknown Input Problem (UIP), focusing on the design of estimators with adaptive characteristics. First, we improve the existing vision-based RSE with an Iterated Extended Kalman Filter (IEKF) to provide better stability, validating the module with a dynamic landing mission. Then, leveraging vision sensors and system dynamics, we integrate the ESO into the IEKF, which enhances the estimator's robustness by concurrently recovering the 6 Degree-of-Freedom (DoF) states and system disturbances. All the proposed modules are validated through simulations and experiments. We also analyze the stability of the employed controllers, estimators, and observers. The main contribution of this thesis is a novel composite control and estimation framework for mobile systems, which promises to enhance the autonomy and reliability of mobile robotics platforms.