Localization - guaranteed navigation in GPS-denied environment via multi-UAV closed-loop coordination

Consider a scenario where multiple Unmanned Aerial Vehicles (UAVs) autonomously collaborate with each other to explore an unknown environment where GPS is not available. A visual fiducial marker with known geometry is fixed on each UAV to provide relative localization between any pair of UAVs. Nevertheless, the UAV s need to plan their motion to ensure that the marker always appears in camera's field-of-view so that they can be localized. Such requirement limits the trajectory space of UAVs when they are exploring the environment. To solve this issue, our first technical contribution is an innovative multi-UAV spatial closed-loop coordination mechanism, which provides guaranteed relative localization wherever they are in the unknown and texture-less environment. The coordination, however, requires that the environment satisfy line-of-sight (LOS) constraints, and therefore necessities the division of the global environment into different subareas such that LOS constraints are met within each subarea. Our second contribution is a novel temporal-spatial pose graph to register different subareas into one global environment accurately. Finally, we present an iterative strategy to simultaneously maximize the volume of exploration space and minimize the localization error under the line-of-sight (LOS) constraints. Comparison with STOA visual localization techniques in simulated unknown environment demonstrates that our method is robust, accurate and independent of the environment.