In recent years, progresses on different aspects of unmanned aerial vehicles (UAVs), especially quadrotors, have promoted an increasing number of applications, including inspection, precision agriculture, and search and rescue. However, to accomplish such tasks efficiently, the capability of high-speed navigation in complex unknown environments is required, which still remains one of the biggest challenges. Besides, the capability of autonomous exploration, in which the vehicle explores and maps the unknown environments to gather information completely and quickly, is also a fundamental component.

In this thesis, we present methods for real-time motion planning, exploration path planning, and multi-robot coordination, enabling quadrotors to autonomously navigate unknown challenging scen...[ Read more ]

In recent years, progresses on different aspects of unmanned aerial vehicles (UAVs), especially quadrotors, have promoted an increasing number of applications, including inspection, precision agriculture, and search and rescue. However, to accomplish such tasks efficiently, the capability of high-speed navigation in complex unknown environments is required, which still remains one of the biggest challenges. Besides, the capability of autonomous exploration, in which the vehicle explores and maps the unknown environments to gather information completely and quickly, is also a fundamental component.

In this thesis, we present methods for real-time motion planning, exploration path planning, and multi-robot coordination, enabling quadrotors to autonomously navigate unknown challenging scenarios at high speeds, as well as explore complex environments efficiently. We start with a kinodynamic path searching and B-spline-based trajectory generation method, which generates a high-quality trajectory in a few milliseconds to support fast and safe flight. We then investigate the local minima issue in trajectory generation and propose a topological path-guided method that thoroughly explores the solution space and improves the quality of generated trajectories. Furthermore, we propose perception-aware planning approaches that enable the quadrotor to perceive and avoid “surprising” obstacles in an active manner, which significantly enhances flight safety in cluttered scenes. After that, we turn to investigating autonomous exploration with one or multiple quadrotors. We start by presenting a hierarchical planning framework that can support fast exploration in complicated unknown environments with a single quadrotor. Based on the above research, collaborative exploration using a fleet of decentralized quadrotors is studied. We present a coordination method that is robust to unstable communication and capable of dispatching the quadrotor team effectively, achieving a much higher exploration rate than a single quadrotor. Throughout the thesis, we conduct extensive benchmark comparisons and challenging real-world experiments, showing the effectiveness of our methods. We open source all our implementations to benefit the community.