Ensuring the safety of multirotors, a prevalent type of aerial robot, during autonomous navigation is paramount, particularly in confined indoor environments where ego airflow disturbances can pose significant hazards. These disturbances can compromise flight stability, resulting in control issues, inaccurate state estimation, and perception problems. Motivated by the need for reliable and safe navigation in such challenging conditions, we focus on Ego-Airflow-Disturbance-Aware Navigation for aerial robots.

In this thesis, we first present the estimation and adaption to indoor airflow disturbances in relatively open areas and integrating these estimates into trajectory planning. Through hover experiments, we model disturbances based on acceleration variance and verify them in complex e...[ Read more ]

Ensuring the safety of multirotors, a prevalent type of aerial robot, during autonomous navigation is paramount, particularly in confined indoor environments where ego airflow disturbances can pose significant hazards. These disturbances can compromise flight stability, resulting in control issues, inaccurate state estimation, and perception problems. Motivated by the need for reliable and safe navigation in such challenging conditions, we focus on Ego-Airflow-Disturbance-Aware Navigation for aerial robots.

In this thesis, we first present the estimation and adaption to indoor airflow disturbances in relatively open areas and integrating these estimates into trajectory planning. Through hover experiments, we model disturbances based on acceleration variance and verify them in complex environments. We employ Hamilton-Jacobi reachability analysis to facilitate safe trajectory planning, combining kinodynamic path search and B-spline trajectory optimization. Our system is validated across multiple multirotor platforms in diverse indoor settings.

Furthermore, we present a comprehensive autonomous aerial system capable of smooth navigation through extremely confined areas, specifically tunnels as narrow as 0.5 m. The flight characteristics are first examined in near-horizontal tunnels through computational fluid dynamics and real flight data analyses. Then, by leveraging the characteristics, a perception-and-disturbance-aware motion planner generating smooth trajectories based on map data and Euclidean distance fields is developed and integrated on the system. It also features a virtual omni-directional perception module and an advanced state estimator, VINS-Multi, optimized for asynchronous multi-camera inputs. Extensive flight experiments in narrow tunnels demonstrate the robustness and superior performance of the system, outperforming even human pilots.

This thesis addresses the safety issues of multirotors in complex environments and extends their flight capability. It also underscores the potential of aerial robots for practical applications in inspection, search and rescue, and other challenging environments.