Real-time Heuristic Framework for Safe Landing of UAVs
Jaskirat Singh*
Neel Adwani*
Harikumar Kandath
K. Madhava Krishna
* denotes equal contribution
To be presented at International Conference on Unmanned Aircraft Systems, Poland
[Paper]
[Video]

Background and Motivation

  • UAVs are susceptible to external events:
    • Loss of communication from ground stations
    • Inability to navigate due to sensor malfunctioning
    • Emergency landing during search and rescue operations
    • Weather disturbances
  • Current studies primarily focus on completely known and static environments, i.e environments that consist of objects at fixed locations.
  • Current frameworks relying on state-of-the-art Deep Learning Models

Proposed RHFSafeUAV Framework Contributions

In order to estimate the seismic structural parameters of the buildings the following modules have been introduced:

  • We developed a framework to land safely in static and dynamic environments where any objects in motion are tracked. As far as we know, no research has been done that specifically addresses this issue in dynamic circumstances where the scenario changes with respect to time.
  • We introduced the dynamic architecture for finding the area of PLZ, and distance of the UAV to the PLZ. The existing literature estimates the resulting area of the potential landing zone using computationally intensive state-of-art deep learning networks.
  • We have compared our results with existing literature showing higher accuracy and precision for area identification with a low computational cost. We evaluated our proposed approach with field tests in real-world dynamic environments.

Results - Potential Landing Zone Detection

  • We validated our PLZ detection module in real-world scenarios. Specifically, we tested the module in Rural, Urban, and Sub-urban scenarios
  • Yellow circles correspond to the identified PLZs, and their resulting diameters (distances) along with the area occupied can be visualized in the Table
  • Time-of-Flight (ToF) sensor for obtaining the ground truth, which has a maximum range of 60 meters.

  • The average percentage error for our distance and area-estimation method is recorded as 0.9692% and 1.9427%
  • Compared the accuracy to that of Google Earth, which has been listed as 1% in [1].
  • Compared our area-estimation resulting average percentage error with the work done by authors that claims for 5% error in [2].

    Results - Real-Time Landing Zone State Estimation

    We studied different scenarios for estimating real-time velocity through flying UAV

  • We validated our algorithm, where we determined the speeds of the moving objects (vehicles), as shown.
  • The ground truth was measured through the vehicles’ odometry. Hence, the results yielded an average percentage error of 2.37%.

    • Trajectory Graph

    UAV Descent Graph

    Acknowledgements

    This template was originally made by Phillip Isola and Richard Zhang for a colorful ECCV project; the code can be found here.

    Contact

    If you have any question, please reach out to any of the above mentioned authors.