SENTINeL: Steps to Enable Near-distance Teamwork under Imperfect Neighbor Localization

Abstract:

Teams of aerial robots could accomplish tasks that a single drone cannot handle as effectively, from transporting objects together to inspecting infrastructure or capturing multiple views of the same scene. But working safely and reliably at close distances requires knowing where neighboring drones are. Current control methods often assume this information is highly accurate, while in real-world conditions, onboard measurements can be noisy, incomplete, and dependent on the robots’ relative positions. The SENTINeL project aims to overcome this limitation by developing control methods that account for uncertainty in how drones locate one another. It will investigate how localization errors affect coordinated flight, develop statistical models of these errors, and design control methods that allow drones to maintain formations despite imperfect information. The methods will be tested in experiments with multiple real drones. The goal is to enable more robust, practical cooperation between aerial robots without relying on external localization infrastructure.
 

Main goal

To create the capability of teams of aerial robots to assume formations despite having imperfect relative localization to each other, and thus to make practicable many impressive feats of robotics that have so-far only been relegated to laboratory demonstrations and very limited field deployment.

Research directions

• Sensor uncertainty analysis and modeling in multi-robot context: How to represent multi-dimensional uncertainty involved in relative localization of cooperating robots in a systeatic, universal and computationally-tractable ways?

• Behavior of robot formations under imperfect relative localization: How does, in general and in specific cases, a formation of robots behave depending on the magnitude, shape and other properties of relative localization uncertainties?

• Uncertainty-resistant formation control: How to design formation-enforcing control algorithms such that they counter the undesirable effects that imperfect relative localization otherwise creates?

In Media: