This paper describes an adaptive method to predict the battery discharge of a multirotor drone over a generic path. A proper assessment of battery state of discharge trend is critical to ensure a safe operation of battery-powered aerial vehicles in critical environments, such as the urban ones. Several standard paths were executed by a commercial, battery-powered drone to acquire the data needed to train a Deep Learning based method; telemetry files and ground-collected data were processed to train the proposed method according to a trajectory segmentation strategy. Two learning configurations were trained to predict the time-of-flight and the integral of the battery current needed to fly the standard path segments. The current integral for each standard path segment is exploited to estimate the corresponding reduction of the battery state of charge. Based on path segmentation into predefined standard sections, the presented solution allows to predict time of flight and battery consumption along any generic path. This can be exploited to plan a complete path at strategic stage, as well as to estimate the remaining available power resource at any intermediate point along a generic trajectory both at strategic stage and during mission execution to be exploited by the drone operator or by a traffic management service. To validate the technique, a scaled package delivery mission is presented as an example of contingency management application. The maximum distance that the drone can reach from selected points along the mission path was computed according to the remaining battery level. Considering a random distribution of safe-landing areas around the planned path, the computed maximum distance was used to determine which safe-landing areas can be reached by the drone from a generic point of the path after an unexpected event.

A data-driven learning method for online prediction of drone battery discharge

Conte C.;Rufino G.;de Alteriis G.;Bottino V.;Accardo D.
2022

Abstract

This paper describes an adaptive method to predict the battery discharge of a multirotor drone over a generic path. A proper assessment of battery state of discharge trend is critical to ensure a safe operation of battery-powered aerial vehicles in critical environments, such as the urban ones. Several standard paths were executed by a commercial, battery-powered drone to acquire the data needed to train a Deep Learning based method; telemetry files and ground-collected data were processed to train the proposed method according to a trajectory segmentation strategy. Two learning configurations were trained to predict the time-of-flight and the integral of the battery current needed to fly the standard path segments. The current integral for each standard path segment is exploited to estimate the corresponding reduction of the battery state of charge. Based on path segmentation into predefined standard sections, the presented solution allows to predict time of flight and battery consumption along any generic path. This can be exploited to plan a complete path at strategic stage, as well as to estimate the remaining available power resource at any intermediate point along a generic trajectory both at strategic stage and during mission execution to be exploited by the drone operator or by a traffic management service. To validate the technique, a scaled package delivery mission is presented as an example of contingency management application. The maximum distance that the drone can reach from selected points along the mission path was computed according to the remaining battery level. Considering a random distribution of safe-landing areas around the planned path, the computed maximum distance was used to determine which safe-landing areas can be reached by the drone from a generic point of the path after an unexpected event.
File in questo prodotto:
Non ci sono file associati a questo prodotto.

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: https://hdl.handle.net/11588/897398
Citazioni
  • ???jsp.display-item.citation.pmc??? ND
  • Scopus 0
  • ???jsp.display-item.citation.isi??? ND
social impact