Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Should I Stay or Should I Go: A Learning Approach for Drone-based Sensing Applications

Giorgos Polychronis, Manos Koutsoubelias, Spyros Lalis

TL;DR

The paper tackles the challenge of minimizing drone sensing mission time when onboard data processing may or may not require immediate action. It introduces a learning-based framework that estimates the time-varying probability of needing action, $p_{wp_i,t}$, from past missions and uses this to decide whether to wait for processing ($d_i=0$) or proceed to the next waypoint ($d_i=1$). The approach combines a formal problem formulation with regression-based probability estimation, memory management, and anomaly detection, and evaluates performance against static policies using both SITL and real-field validation. Results show up to 25.8% mission-time reduction and robust adaptation to changing environments, with a decision-tree regressor providing the best accuracy among tested predictors. The work has practical implications for accelerating autonomous drone sensing tasks and can be extended to dynamic or multi-drone scenarios.

Abstract

Multicopter drones are becoming a key platform in several application domains, enabling precise on-the-spot sensing and/or actuation. We focus on the case where the drone must process the sensor data in order to decide, depending on the outcome, whether it needs to perform some additional action, e.g., more accurate sensing or some form of actuation. On the one hand, waiting for the computation to complete may waste time, if it turns out that no further action is needed. On the other hand, if the drone starts moving toward the next point of interest before the computation ends, it may need to return back to the previous point, if some action needs to be taken. In this paper, we propose a learning approach that enables the drone to take informed decisions about whether to wait for the result of the computation (or not), based on past experience gathered from previous missions. Through an extensive evaluation, we show that the proposed approach, when properly configured, outperforms several static policies, up to 25.8%, over a wide variety of different scenarios where the probability of some action being required at a given point of interest remains stable as well as for scenarios where this probability varies in time.

Should I Stay or Should I Go: A Learning Approach for Drone-based Sensing Applications

TL;DR

The paper tackles the challenge of minimizing drone sensing mission time when onboard data processing may or may not require immediate action. It introduces a learning-based framework that estimates the time-varying probability of needing action, , from past missions and uses this to decide whether to wait for processing () or proceed to the next waypoint (). The approach combines a formal problem formulation with regression-based probability estimation, memory management, and anomaly detection, and evaluates performance against static policies using both SITL and real-field validation. Results show up to 25.8% mission-time reduction and robust adaptation to changing environments, with a decision-tree regressor providing the best accuracy among tested predictors. The work has practical implications for accelerating autonomous drone sensing tasks and can be extended to dynamic or multi-drone scenarios.

Abstract

Multicopter drones are becoming a key platform in several application domains, enabling precise on-the-spot sensing and/or actuation. We focus on the case where the drone must process the sensor data in order to decide, depending on the outcome, whether it needs to perform some additional action, e.g., more accurate sensing or some form of actuation. On the one hand, waiting for the computation to complete may waste time, if it turns out that no further action is needed. On the other hand, if the drone starts moving toward the next point of interest before the computation ends, it may need to return back to the previous point, if some action needs to be taken. In this paper, we propose a learning approach that enables the drone to take informed decisions about whether to wait for the result of the computation (or not), based on past experience gathered from previous missions. Through an extensive evaluation, we show that the proposed approach, when properly configured, outperforms several static policies, up to 25.8%, over a wide variety of different scenarios where the probability of some action being required at a given point of interest remains stable as well as for scenarios where this probability varies in time.
Paper Structure (23 sections, 5 equations, 8 figures)

This paper contains 23 sections, 5 equations, 8 figures.

Table of Contents

  1. Introduction
  2. Related Work
  3. Problem Formulation
  4. Missions
  5. Flight model
  6. Decision, time gain and time penalty
  7. Mission time
  8. Approach
  9. Learning from past experience
  10. Experience memory size and reset
  11. Drone control logic
  12. Validation of High-level Simulator
  13. Validation of SITL vs real flight behavior
  14. Validation of high-level simulator vs SITL
  15. Evaluation
  16. ...and 8 more sections

Figures (8)

  • Figure 1: Drone in the field vs SITL configuration.
  • Figure 2: Timeline of a complex mission execution scenario using the autopilot in SITL mode vs the high-level simulator.
  • Figure 3: Mission area and path plan.
  • Figure 4: Experience memory comparison. Stable world. (a) out scenario, (b) out-in scenario, (c) in-out scenario.
  • Figure 5: Experience memory comparison. Changing world. (a) out scenario, (b) out-in scenario, (c) in-out scenario.
  • ...and 3 more figures