Vision Object-Oriented Augmented Sampling-Based Autonomous Navigation for Micro Aerial Vehicles

Abstract

Autonomous navigation of micro aerial vehicles in unknown environments not only requires exploring their time-varying surroundings, but also ensuring the complete safety of flights at all times. The current research addresses estimation of the potential exploration value neglect of safety issues, especially in situations with a cluttered environment and no prior knowledge. To address this issue, we propose a vision object-oriented autonomous navigation method for environment exploration, which develops a B-spline function-based local trajectory re-planning algorithm by extracting spatial-structure information and selecting temporary target points. The proposed method is evaluated in a variety of cluttered environments, such as forests, building areas, and mines. The experimental results show that the proposed autonomous navigation system can effectively complete the global trajectory, during which an appropriate safe distance could always be maintained from multiple obstacles in the environment.

UAV Control in Autonomous Object-Goal Navigation: A Systematic Literature Review

“…Other methods are proposed that use genetic (Tymochko et al, 2020;Song et al, 2022) or optimization (Delamer et al, 2021;Jia et al, 2022) algorithms. An interesting approach was presented by Zhao et al (2021b) with intermediate goals to reach the final known point. The custom implementations were most mathematical models using position information on the UAV and the target location, planning the pose, velocity, or future state that must be reached.…”

“…As shown in Table 10, nine papers treat the mapping task as reconstruction, fusing sensory data to create a spatial projection for each element in the scene. Simultaneous mapping and localization (SLAM) (Bachrach et al, 2010;Bershadsky and Johnson, 2013;Dang et al, 2020;Zhao et al, 2021b), a widely used method in navigation, solves this reconstruction problem. SLAM fuses internal and external sensor information to create a map with structural information such as corridors, objects, and the vehicle itself.…”

“…A variation of occupancy maps adds cell priority (Kusnur et al, 2021) according to a computed score or adds OctoMap (Dang et al, 2018), which includes the Octree data structure for efficiency. Other variations were implemented with SLAM for sensory data fusing (Dang et al, 2020;Zhao et al, 2021b). A final category for the mapping task is the reference map, in which the methods use environment information as a reference for spatial information.…”

“…One category was named probabilistic, with seventeen articles using this kind of method, processing the vehicle sensor data with probability techniques to infer an expected state. In this category was included the SLAM technique (Bachrach et al, 2010;Bershadsky and Johnson, 2013;Cocchioni et al, 2014;Dang et al, 2020;Narazaki et al, 2022;Kumar et al, 2020;Zhao et al, 2021b) because the sensor data is processed to correct the current vehicle pose. Additionally, SLAM is complemented with probability methods such as Kalman filters with its variations of extended Kalman filters (EKF) (Konrad et al, 2017;LI and HU, 2021;Mansouri et al, 2020b;Meng et al, 2019) and Kalman consensus filter (Li et al, 2018), non-linear observer (Helgesen et al, 2019), particle filters (Wang et al, 2017), and Monte Carlo processes (Delamer et al, 2021;Kusnur et al, 2021).…”

“…Another custom implementations use Density-Based Spatial Clustering of Applications with Noise (DBSCAN) approach (Chen and Lu, 2022;Prophet and Trommer, 2020) to produce surrounding information from laser data. Another method processes the sensor data a spatial-structure information (Zhao et al, 2021b) to solve the autonomous navigation problem.…”