Adaptation is the Key: Scalable Robot Swarms & Robust Autonomy

Presenter: Souma Chowdhury, PhD, University at Buffalo
Location: 1043 ERF and on Zoom at https://uic.zoom.us/j/89524344993?pwd=MnIwdXRlRGN1cWFKcEVCRDZBcVJNQT09
Meeting ID: 895 2434 4993
Passcode: uicmie123

Abstract: Adaptation is fundamental to the performance of robotic and autonomous systems (RAS) that operate in complex, unstructured and uncertain environments, with nature-inspired algorithms, optimization and machine learning forming important pillars towards accomplishing effective adaptation. Within this context, this seminar talk will touch upon our work in two major areas: 1) One where large teams of robots must coordinate to perform critical operations such as disaster response, with environment variations, adversities, limited onboard computing and need for scalability with team and task size presenting key barriers to effective coordination. 2) Another where measuring the behavior or impact of individual RAS, e.g., unmanned aerial vehicles (UAVs), under uncertainties is expensive, thereby limiting the availability of data to train prediction models and design architectures for planning and control of these RAS.

In the area of multi-robot systems and robot swarms, we have developed a suite of algorithms to guide the collective behavior of multiple ground or aerial robots for operations such as signal source localization, multi-location response, area coverage and boundary mapping. These algorithms are built using a variety of methods such as optimization, graph theory and metaheuristics. We have also explored their application in simulated use cases such as flood response, finding skiers trapped under avalanche and offshore oil spill mapping. More recently, we have broken new ground on scalability by innovating graph learning approaches to generate real-time policies for multi-robot task allocation and vehicle routing. A culmination of our work in swarm robotics has been the demonstration of new reinforcement learning frameworks and gaming environments (for human-swarm interaction) to model situation-adaptive tactical decisions for teams of UAVs and UGVs performing victim search in complex urban environments.

In the problem area of modeling the behavior of RAS, our work in recent years has focused on the emerging paradigm of physics-infused machine learning (PIML). More specifically, we have introduced new hybrid architectures that combine neural nets (e.g., MLP, TCN and LSTM) with simplified physics models to closely match high-fidelity data, while providing substantially better generalization and potentially improved explainability, compared to purely black-box neural net models. Applications of PIML have included modeling the trajectory of quadcopters under gusts, mitigating draft interaction between two quadcopters, and modeling the 3D noise field of quadcopters in indoor environments.

Speaker Bio: Souma Chowdhury is an associate professor of Mechanical and Aerospace Engineering at University at Buffalo, where he leads the Adaptive Design Algorithms, Models and Systems (ADAMS) Lab. Chowdhury graduated with his Ph.D. in mechanical engineering from Rensselaer Polytechnic Institute in Troy, NY. Prior to joining the University at Buffalo, he worked as a post-doctoral researcher and research faculty at Syracuse University and Mississippi State University. His research interests lie at the intersections of multi-fidelity optimization, evolutionary computing and machine learning, with applications to design and control of autonomous systems, swarm robotics, design of metamaterial systems and resilience of infrastructural networks. He has co-authored 44 peer-reviewed journal articles, 95 full-length conference articles, and two book chapters in related topics. His research work has been supported by funding from NSF, DARPA, ONR and AFOSR, including the NSF CAREER Award. He is a member of the ASME, AIAA, and IEEE societies, and a member of the AIAA Multidisciplinary Design Optimization Technical Committee, where he chairs the Education Sub-Committee. He is also the co-founder and coordinator of the symposiums on AI for Design and Systems Science and Design of Autonomous Systems, at the ASME IDETC conferences. He has received the 2019 SEAS Early Career Researcher of the Year award at the University at Buffalo, and the Best Paper award at the 2021 IEEE MRS conference for his paper on AI-based swarm tactics.