References of "Annaiyan, Arun 50000421"
     in
Bookmark and Share    
Full Text
See detailCollision-Free Navigation of Small UAVs in Complex Urban Environment
Annaiyan, Arun UL

Doctoral thesis (2018)

Small unmanned aerial vehicles (UAVs) are expected to become highly innovative solutions for all kind of tasks such as transport, surveillance, inspection or guidance, and many commercial ideas already ... [more ▼]

Small unmanned aerial vehicles (UAVs) are expected to become highly innovative solutions for all kind of tasks such as transport, surveillance, inspection or guidance, and many commercial ideas already exist. Herein, small multi rotor UAVs are preferred since they are easy to construct and to fly, at least in wide open spaces. However, many UAV business cases are foreseen in complex urban environments which are very challenging from the perspective of UAV flight. Our work focuses on the autonomous flight and collision-free navigation in an urban environment, where GPS is still considered for localization but where variations in the accuracy or temporary unavailability of GPS position data is explicitly considered. Herein, urban environments are challenging because they require flight nearby large structures and also nearby moving obstacles such as humans and other moving objects, at low altitudes or in very narrow spaces and thus also in areas where GPS (global positioning system) position data might temporarily be very inaccurate or even not available. Therefore we designed a custom stereo camera with adjustable base length for the perception of the possible potential obstacles in the unknown outdoor environment. In this context the optimal design and sensitivity parameters are investigated in outdoor experiments. Using the stereo images, graph based SLAM approach is used for online three dimensional mapping of the static and dynamic environment. For the memory efficiency incremental online loop closure detection using bag of words method is implemented here. By having the three dimensional map, the cost of the cell and its transition calculated in real time by the modified D* lite which will search and generate three dimensional collision free path planning. Experiments of the 3D mapping and collision free path planning are conducted using small UAV in outdoor scenario. The combined experimental results of real time mapping and path planning demonstrated that the three dimensional collision free path planning is able to handle the real time computational constraints while maintaining safety distance. [less ▲]

Detailed reference viewed: 130 (12 UL)
Full Text
Peer Reviewed
See detailReal-time graph-based SLAM in unknown environments using a small UAV
Annaiyan, Arun UL; Olivares Mendez, Miguel Angel UL; Voos, Holger UL

in 2017 International Conference on Unmanned Aircraft Systems (ICUAS); Miami 13-16 June 2017 (2017)

Autonomous navigation of small Unmanned Aerial Vehicles (UAVs) in cluttered environments is still a challenging problem. In this work, we present an approach based on graph slam and loop closure detection ... [more ▼]

Autonomous navigation of small Unmanned Aerial Vehicles (UAVs) in cluttered environments is still a challenging problem. In this work, we present an approach based on graph slam and loop closure detection for online mapping of unknown outdoor environments using a small UAV. Here, we used an onboard front facing stereo camera as the primary sensor. The data extracted by the cameras are used by the graph-based slam algorithm to estimate the position and create the graph-nodes and construct the map. To avoid multiple detections of one object as different objects and to identify re-visited locations, a loop closure detection is applied with optimization algorithm using the g2o toolbox to minimize the error. Furthermore, 3D occupancy map is used to represent the environment. This technique is used to save memory and computational time for the online processing. Real experiments are conducted in outdoor cluttered and open field environments.The experiment results show that our presented approach works under real time constraints, with an average time to process the nodes of the 3D map is 17.79ms. [less ▲]

Detailed reference viewed: 248 (19 UL)