Robotic Sensor Networks Lab - NRI: Targeted Observation of Severe Local Storms Using Aerial Robots

NRI: Targeted Observation of Severe Local Storms Using Aerial Robots

Introduction

As part of NSF 1526200, our group has been working on general capabilities for tracking signal sources and energy efficient coverage.

Signal Source Localization

Consider a signal source (e.g. a radio transmitter - VHF collar) to be localized. The goal is to minimize the time spent in localizing the source in such a way that the localization uncertainty is below a desired level.

Approach-I - Online Localization with Single UAV

Let the true target location be denoted by w. The robot can take bearing measurements of the target location. Each bearing measurement yields a bounded wedge with angular clearance such that the wedge will contain the target. The localization of the target is performed by intersecting the wedges obtained from bearing measurements. Each measurement requires a fixed amount of time. Locations from which measurements are taken are denoted by s_i for a given measurement i. The full sequence of N measurement locations is S = {s₁, ..., s_n} . The problem is to determine measurement locations s_i in an online manner so as to localize a target to a desired localization uncertainty U* such that the time spent in taking measurements and traveling will be minimized.
In the figure, the UAV is moving from s_i, to s_i+1 to take the next bearing measurement. The localization is done by intersecting the bearing measurements. The uncertainty is a function of the measurement locations relative to the (unknown) target location w.
We propose an online algorithm to locate a target using bounded bearing measurements along with a cost analysis of the algorithm, and implement the algorithm on a quadrotor to localize a VHF collar, along with its validation in field experiments.

Approach-II - Multi-UAV System for Signal Source Localization

UAVs with Yagi antenna in the take-off area

Given an environment T, compute the measurement locations S = {s₁, ..., s_n} and the tour visiting them for each UAV such that the following criteria are satisfied: (i) the time spent in visiting these locations and taking measurements is minimized, (ii) the resulting localization uncertainty U for each location in T is below the desired uncertainty U*, (iii) the collision avoidance among UAVs is guaranteed.
At a high-level, the data gathering strategy is to compute measurement locations along with tours to visit them. Each tour is assigned to a team of three robots. The teams operate independently. The three robots in each team execute a coordinated strategy in which the leader signals when to navigate to the next measurement location, take measurements, and when to go home in case of emergency.
We validated the system in a series of field experiments using three autonomous aerial robots equipped with an on-board computer, directional antenna, and signal receiver.
It was shown that the coordinated motion outperforms the non-coordinated motion in terms of the localization quality when the target is moving.

Media

icra2018-video-muted.mp4

mrs2017-video-muted.mp4

Energy-aware Coverage

We developed the coverage path planning algorithm when the robot has limited energy budgets .

Given:

(1) An environment to cover.

(2) The robot's energy budget (the maximum distance the robot could move after a full recharge)

Goal:

Find a set of paths, which fully covers the given environment and minimize the path number.

videoSub_finalV3_compressed.mp4

Result 1: An approximation algorithm whose performance is bounded by 4OPT for contour-connected environments, and 4(2r+1)OPT for any environments, where r is the number reflex vertices of the environments.

Result 2: An approximation algorithm whose performance is bounded by 4log(A) of the optimal solution, where A is the size of the environment.

The field experiment at Itasca State Park using result 1.

The field experiment at Itasca State Park using result 2.

Publications

H. Bayram, N. Stefas, V. Isler, “Aerial Radio-based Telemetry for Tracking Wildlife“, IEEE International Conference on Robotics and Automation (ICRA), Australia, 2018 (submitted).
H. Bayram, N. Stefas, K.S. Engin, V. Isler, “Tracking Wildlife with Multiple UAVs: System Design, Safety and Field Experiments“, IEEE International Symposium on Multi-robot and Multi-agent Systems (MRS), Los Angeles, USA, 2017 (to appear).
H. Bayram, K. Doddapaneni, N. Stefas,, V. Isler, “Active Localization of VHF Collared Animals with Aerial Robots“, IEEE International Conference on Automation Science and Engineering (CASE), Texas, USA, pp. 934-939, 2016.
M. Wei and V. Isler, "Coverage Path Planning under the Energy Constraint", IEEE International Conference on Robotics and Automation, 2018.
M. Wei and V. Isler, "A Log-Approximation for Coverage Path Planning with the Energy constraint", The International Conference on Automated Planning and Scheduling, 2018.
M. Wei and V. Isler, "Approximation Algorithms for Coverage Path Planning with the Energy Constraint", The International Journal of Robotics Research. (submitted)

Acknowledgement

This material is based upon work supported by the National Science Foundation under Grant No. 1525045

Any opinions, findings, and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the National Science Foundation.

UAV Radio-based Autonomous Landing

Figure 1 : Cedar Creek Ecosystem Science Reserve field test

Figure 2 : ND region approximation and measurement locations for the Cedar Creek field test

Consider the task of landing on an area around a target signal source (e.g. VHF radio transmitter). The goal is to minimize the total flight time required to approach and land on the desired target area.
We show that there exists a region above the target inside of which bearing measurements lose directionality : signal recordings in all directions yield similar signal strength. We name this region "No Directionality" (ND) region and we can detect whether a location is inside the region using short-time ND binary measurements. We approximate this region with a cone.
We propose an online strategy to approach and land on the desired area that takes advantage of a UAVs ability to change height and exploit the structure of the ND region (occurring when approaching the target beacon from above) in order to significantly reduce flight time.
We test our approach on a multi-rotor UAV at the Cedar Creek Ecosystem Science Reserve (see Figures). The UAV takes an initial bearing measurement to determine direction of the target area. Then it increases height and only takes short binary measurements until it has entered the ND region. Finally, it reduces altitude and re-enters the ND region until it's size is the desired landing area size.

Report abuse