We invent a manual control program that works regardless of delay and is as intuitive as possible.

To solve the delay problem we thought of using a machine learning algorithm and will interpolate its data using Kalman filters to correct and predict the trajectory according to the driver input. The position of the drone will need to be constantly monitored to ensure, of course, that it maintains its established route.

Kalman filters work on two variables, both known by the drone algorithm: its last position since the autonomous systems have failed and the drone's speed and speed.

https://upload.wikimedia.org/wikipedia/commons/a/a5/Basic_concept_of_Kalman_filtering.svg

The machine learning algorithm generates a route that returns the drone to the correct path in a generative way, in a closed loop system using several simulations of the trajectory error made at a shorter time to increase the efficiency and accuracy of the algorithm (in worst case a second).

The driver controls the drone by moving the phone on certain axes, it sees the simulation of the trajectories on a screen, at the same time using the sensors with which it is equipped to keep away from obstacles - demonstration.

If the drone has physical defects, it will fail in a plateau area where it will send bacon signals to be recovered.

Links to our code

https://github.com/GhiaraD/DroneApp

https://github.com/GhiaraD/ML-for-routes

Resources:

https://www.nasa.gov/press-release/nasas-dragonfly-will-fly-around-titan-looking-for-origins-signs-of-life

https://www.nasa.gov/uamgc

https://dev.socrata.com/

https://www.nasa.gov/centers/dryden/pdf/175940main_Earth_Obs_UAV_Vol_2_v1.1_Final.pdf

https://en.wikipedia.org/wiki/PID_controller

https://www.sesarju.eu/sites/default/files/documents/sid/2017/SIDs_2017_paper_65.pdf