Session WISARN-O


8:50 AM — 9:00 AM EDT
May 20 Sat, 8:50 AM — 9:00 AM EDT

Session Chair

Marco Di Felice (University of Bologna)

Session WISARN-K1


9:00 AM — 10:00 AM EDT
May 20 Sat, 9:00 AM — 10:00 AM EDT

Multi-Robot Coordination

Rernhard Rinner (Institute of Networked and Embedded Systems, University of Klagenfurt, Austria)

Now that robots have evolved from bulky platforms to agile devices, a challenge is to combine multiple robots into an integrated autonomous system, offering functionality that individual robots cannot achieve. A key building block for this integration is coordination, which is concerned with sharing knowledge, joint decision making, and allocation of computation tasks to processing nodes. Different levels of coordination exist—ranging from high-level functions, like the assignment of system-wide tasks and resources, down to low-level control, like collision avoidance, flight formations, and joint sensor usage for state estimation.
In this talk, I will present different coordination techniques for multi-robot systems, discuss their usage in prototypical multi-robot applications such as path planning, swarming and surveillance, and demonstrate deployments in unmanned aerial vehicles.
Speaker Rernhard Rinner (Institute of Networked and Embedded Systems, University of Klagenfurt, Austria)

Session Chair

Marco Di Felice (University of Bologna)


Virtual Coffee Break

10:00 AM — 10:30 AM EDT
May 20 Sat, 10:00 AM — 10:30 AM EDT

Session WISARN-S1

Technical Session

10:30 AM — 12:00 PM EDT
May 20 Sat, 10:30 AM — 12:00 PM EDT

Modeling a System of Interconnected Quadrotor UAVs for Suspended Payload Transportation

Özhan Bingöl (Gumushane University, Turkey); Hacı Mehmet Güzey (Sivas University of Science and Technology, Turkey)

Load transportation applications using quadrotors have become a major topic because of the current rise in interest in quadrotor research. There may be various advantages to using more than one quadrotor in a given application, such as boosting cargo capacity or safety, due to their limited size and capacity. But it stands to reason that systems with multiple quadrotors would provide far greater control difficulties than those with a single quadrotor. Additionally, designing an effective controller structure may be rather difficult when taking into account external disturbances and parameter uncertainty. A new neural network-based sliding mode controller (NSMC) for networked quadrotor UAVs carrying a suspended payload is proposed in this paper to overcome such challenges. The neural network component avoids chattering effects in the SMC's control signals and boosts the SMC's effectiveness against time-varying dynamic uncertainties, even if the suggested controller makes use of the resilient structure of the SMC for the nonlinear system. An interconnected quadrotor unmanned aerial vehicle (UAV) dynamics formulation is first built for this purpose. The suggested controller is then created after that. Once the controllers are in place, the stability of the system is shown using the Lyapunov stability criterion, and the effectiveness of the suggested controller is validated using numerical simulations.
Speaker biography is not available.

Semantic and Effective Communication for Remote Control Tasks with Dynamic Feature Compression

Pietro Talli, Francesco Pase and Federico Chiariotti (University of Padova, Italy); Andrea Zanella (University of Padova, Italy & CNIT, Italy); Michele Zorzi (University of Padova, Italy)

The coordination of robotic swarms and the remote wireless control of industrial systems is one of the major use cases for 6G: in these cases, the massive amounts of sensory information that needs to be shared over the wireless medium can overload even high-capacity connections. In these cases, solving the effective communication problem by optimizing the transmission strategy to discard irrelevant information can provide a significant advantage, but is often a very complex task. In this work, we define a remote Partially Observable Markov Decision Process (POMDP) model in which an observer must communicate its sensory data to an actor controlling a task (e.g., a mobile robot in a factory), and consider the semantic and effective communication problems. We split the task by considering an ensemble Vector Quantized Variational Autoencoder (VQ-VAE) encoding and train a Deep Reinforcement Learning (DRL) agent to dynamically adapt the quantization level, considering both the current state of the environment and the memory of past messages. Our simulations show that this scheme obtains a significant performance increase on the classic Cartpole problem.
Speaker Pietro Talli (University of Padova)

I received my Master Degree in ICT for Internet and Multimedia Engineering (University of Padova) in 2022. I am a PhD student at the SIGNET research group at the University of Padova. I am currently working on a PNRR project focusing on communication systems in extreme environments. My research interests includes also new communication paradigm such as Semantic and Effective Communication.

A Hasty Grid S&R Prototype Using Autonomous UTM and AI-Based Mission Coordination

Lanier Watkins and Denzel Hamilton (JHU Applied Physics Lab); Chad Mello (United States Air Force Academy, USA); Tyler Young and Sebastian Zanlongo (Johns Hopkins University Applied Physics Lab, USA); Barbara Kobzik-Juul (JHU Applied Physics Lab, USA); Randall Sleight (Johns Hopkins University Applied Physics Lab, USA)

Search and Rescue (S&R) is an old and very important problem. Generally, it is critical to find a missing person in the first few hours; more specifically 86% of battlefield deaths happen a half-hour after injury. Consequently, federal, state, local governments, and industry are highly interested in quickly finding missing persons, injured soldiers and pilots, or disaster victims in hostile environments. We offer a new S&R approach based on the use of autonomous drones or unmanned aircraft systems (UAS) Traffic Management (UTM) and melding two competing approaches, Hasty and Grid S&R. Hasty S&R uses minimum resources to quickly search areas frequented by targets, and Grid (S&R) uses maximum resources to exhaustively search an entire area. Our approach involves using an autonomous UTM system to safely pack enough mission coordinated autonomous drones (i.e., resources) into airspace such that the entire ground below can be BOTH quickly and thoroughly searched. Our results demonstrate the feasibility of this approach in realistic simulated environments with varying weather, sensor visibility, obstacle density, terrain roughness, and battery levels. Further, we built working prototypes that allowed our MATLAB UTM simulation to interact with ArduPilot Software-In-The-Loop simulated drones and real hardware drones that recognized targets using Convolutional Neural Networks. Also, these prototypes demonstrated the ability to identify targets and to coordinate S&R across multiple drones
Speaker Lanier Watkins (Johns Hopkins University)

Lanier Watkins, Chair of the Johns Hopkins University Engineering for Professionals Master’s in Computer Science and Cybersecurity programs, develops innovative algorithms and frameworks to address the continuously changing needs of defending Critical Infrastructure (CI) networks and systems.

 His research efforts are concentrated in five areas: 1) network security, namely introducing new covert channels, cloud paradigms, and network-based detectors to produce both offensive and defensive capabilities; 2) Internet of Things security, with a focus on mobile, cyber-physical, and wireless sensor/medical device security; 3) vulnerability monitoring and analysis, introducing new risk management and security assessment frameworks for IoT devices; 4) malware monitoring and analysis, exploring active malware defenses to contribute to the increasingly popular “hacking back” paradigm; and 5) data analytics and measured artificial intelligence, investigating the use of autonomous decision-making and methods of AI assurance and security to help data scientists and engineers defend CI against traditional threats and the inevitable threat of adversarial AI.

 In addition to advising, lecturing, and mentoring for and Chairing the EP Computer Science and Cybersecurity Master’s programs, Watkins is Principal Staff and a section supervisor in the Critical Infrastructure Protection Group within the Asymmetric Operations Sector of the Johns Hopkins University Applied Physics Laboratory. He also holds a secondary appointment as an associate research professor with the JHU Information Security Institute. Prior to joining APL, Watkins worked for over ten years in industry, first at the Ford Motor Company and later at AT&T.

 Among his awards are Black Engineer of the Year’s Modern-Day Technology Leader Award 2015, as well as APL’s Lawrence R. Hafstad Fellowship 2016 - Present. A senior member of the Institute of Electrical and Electronics Engineers and a member of the Association for Computing Machinery, Watkins has published more than 50 conference papers, journals, and book chapters, and holds several patents and provisional patents for Android mobile device monitoring systems and drone counter defense.

 He received his BS and MS in physics and MS in computer science from Clark Atlanta University, his MS in biotechnology from Johns Hopkins University, and his PhD in computer science from Georgia State University.

Communication-Efficient Reinforcement Learning in Swarm Robotic Networks for Maze Exploration

Ehsan Latif, WenZhan Song and Ramviyas Parasuraman (University of Georgia, USA)

Smooth coordination within a swarm robotic system is essential for the effective execution of collective robot missions. Having efficient communication is key to the successful coordination of swarm robots. This paper proposes a new communication-efficient decentralized cooperative reinforcement learning algorithm for coordinating swarm robots. It is made efficient by hierarchically building on the use of local information exchanges. We consider a case study application of maze solving through cooperation among a group of robots, where the time and costs are minimized while avoiding inter-robot collisions and path overlaps during exploration. With a solid theoretical basis, we extensively analyze the algorithm with realistic CORE network simulations and evaluate it against state-of-the-art solutions in terms of maze coverage percentage and efficiency under communication-degraded environments. The results demonstrate significantly higher coverage accuracy and efficiency while reducing costs and overlaps even in high packet loss and low communication range scenarios.
Speaker Ehsan Latif (University of Georgia)

I am a doctoral candidate at the UGA School of Computing and Research Robotics. I am actively conducting research in multi-robotic systems to improve localization accuracy in dense and dynamic environments and efficient exploration strategies to reduce computation and communication overheads. I am working as a graduate research assistant under the supervision of Dr. Ramviyas Nattanmai Parasuraman in the Heterogeneous Robotics Lab at the University of Georgia Computer Science Department. My research in the HeRo lab involves robotics localization and exploration algorithms using unconventional sensing modalities in unstructured dense, dynamic environments. I am enthusiastic about working with physical as well as simulation of robotic systems, also have experience in software development solutions (ROS, Python).

Attention-Guided Synthetic Data Augmentation for Drone-based Wildfire Detection

Julia Boone, Bryce Hopkins and Fatemeh Afghah (Clemson University, USA)

Drone-based wildfire detection models allow for real-time fire monitoring which is critical for the most efficient intervention and mitigation techniques needed for wildfires. However, due to restrictions on the usage of UAVs during wildfires and prescribed burns, current UAV-sourced wildfire imagery datasets are limited to images of individual burns. While deep learning fire detection models trained and tested on these datasets can achieve high fire classification accuracy, these models fail to generalize when given images of wildfires from other forestry types due to the differences in vegetation, climate, time of year, and other factors that contribute to the visual appearance of these burns. Synthetic augmentation techniques can increase the diversity of training datasets. In this work, we develop an attention-guided image-to-image translation tool that utilizes Generative Adversarial Networks (GANs) to generate wildfire images from aerial forestry images in order to increase classification accuracy. We illustrate the need for attention mechanisms for generating wildfire images through image-to-image translation techniques. We observe increased classification accuracy for a test set based on a separate burn from the training dataset when augmenting training data with diverse synthetic wildfire images.
Speaker Julia Boone

Julia Boone ([email protected]) received her B.S. degree in Computer Engineering from Clemson University in 2022. She is pursuing her Ph.D. degree in Computer Engineering with a focus area of Intelligent Systems at Clemson University. Julia is currently working as a graduate research assistant under Dr. Fatemeh Afghah ([email protected]) in the Intelligent Systems and Wireless Networking (IS-WiN) Laboratory at Clemson University. Her current research interests include generative adversarial networks, trust-monitoring in multi-agent systems, and decision making in multi-agent systems.

Session Chair

Carlos Kamienski (Federal University of ABC, Brazil)

Gold Sponsor

Gold Sponsor

Bronze Sponsor

Student Travel Grants

Student Travel Grants

Local Organizer

Made with in Toronto · Privacy Policy · INFOCOM 2020 · INFOCOM 2021 · INFOCOM 2022 · © 2023 Duetone Corp.