Trajectory generation and tracking control for fast autonomous tail-sitter UAVs

Abstract

This thesis addresses critical challenges in the development of autonomous tail-sitter unmanned aerial vehicles (UAVs), with a focus on trajectory generation and tracking control. Tail-sitter UAVs are a type of hybrid fixed-wing and rotary-wing aircraft that is of exceptional efficiency and maneuverability, ideal for multiple missions in practice. Fast autonomous tail-sitter UAVs capable of high-speed navigation in cluttered real-world environments, hold substantial socio-economic value across diverse applications. Yet, these tail-sitter UAVs pose significant technical challenges. Their aggressive maneuvers involve large variations in both system states and inputs, necessitating a global tracking controller to regulate an underactuated system with highly nonlinear aerodynamics, and a trajectory generation framework to efficiently compute safe and feasible trajectory commands for precise control. From both theoretical and practical perspectives, this thesis contains several key contributions as follows:

1. On-manifold Model Predictive Control (MPC): To achieve high-accuracy trajectory tracking for full-envelope aggressive UAV flights, we develop an on-manifold MPC, bridging the gap between advanced controls typically for linear systems and nonlinear robotic systems subject to curved manifold constraints. This approach is also generalized into a unified framework for common robotic systems on manifolds. By exploiting the differential geometry, the proposed on-manifold MPC features minimal parameterization and avoidance of singularities, demonstrating advantages in aggressive global trajectory tracking.

2. Trajectory Generation and Tracking Control for Tail-Sitters}: Building on the on-manifold MPC, we systematically address both the theoretical and practical problems on trajectory generation and global tracking control for aggressive tail-sitter flights. We prove the differential flatness property of tail-sitter UAVs, fully exploiting realistic aerodynamics models without simplifications. By leveraging this fundamental property, we develop a trajectory optimization to generate dynamically feasible trajectories for aggressive flights, and a global control framework to ensure high tracking accuracy throughout the entire flight envelop. Practical issues of environmental wind effects and actuator saturation are also thoroughly incorporated into both the planning and control framework.

3. Autonomous Tail-Sitter UAVs}: To further enhance the tail-sitter autonomy to autonomous obstacle avoidance, we incorporate obstacle avoidance into the tail-sitter trajectory generation, and effectively improve the computational efficiency to solve the optimization. We develop an efficient, feasibility-first nonlinear programming (NLP) solver, tailored for real-time planning. Integrating with state-of-the-art perception, planning, and control algorithms, we present a fully autonomous tail-sitter UAV capable of high-speed navigation in unknown, cluttered, real-world environments.

Throughout this thesis, theoretical proofs, statistical analysis and extensive experimental validations have been conducted, aiming to advance both theories and technologies for robotic systems applicable to practical uses.