Title: Learning Collaboration in Large-Scale Decentralized Robotic Systems
Date and time: 9 May 2025 (Friday), 9:30 – 10:30 a.m.
Venue: IEOR Seminar Room
Speaker: Dr. Saurav Agarwal, University of Pennsylvania
Abstract:
Decentralized and Collaborative Intelligent Systems (DCIS) offer resilient, scalable, and efficient frameworks for coordinating large teams of autonomous robots. This talk presents a learning-based approach to enabling such systems, where robots collaborate through local perception, selective inter-robot communication, and distributed decision-making. We focus on learning over dynamic communication graphs to determine what information should be shared among neighboring agents and how to act upon received information to achieve global objectives. The talk introduces architectural advances in graph neural networks, transformer-based models, and diffusion policies tailored for decentralized settings. We also present theoretical insights into scalability and generalization in large multi-agent systems, along with results from real-world deployments.
Bio:
Saurav Agarwal is a postdoctoral researcher at the University of Pennsylvania, working with Professors Vijay Kumar and Alejandro Ribeiro. His research focuses on the development of theoretical foundations, algorithm design, and learning-based methods for large-scale decentralized and collaborative intelligent robotic systems. His interests span learning on graphs, combinatorial optimization, and approximation algorithms, with the goal of building efficient, adaptive, and resilient multi-agent systems capable of online decision-making. He received his Ph.D. in Computer Science from the University of North Carolina at Charlotte, where he introduced the generalized coverage problem for monitoring complex environments, developed scalable algorithms, and validated them through deployments involving UAVs and ground robots.