Efficient and Intelligent Collaborative Autonomy in Resource-Constrained, Unpredictable, and Untrusted Environments: A Submodular Optimization and Online Learning Perspective

March 01, 2024, ESB 2001

Vasileios Tzoumas

Michigan, Aerospace


I will present coordination algorithms for distributed artificial intelligence at scale despite severe resource constraints and limited information about the environment. Resource constraints include limited communication speed; and limited information occurs in unpredictable environments such as in target tracking where the targets’ motion model is unknown. This research is motivated by the engineering vision for an autonomy of ubiquitously deployed robots that can collaboratively execute complex tasks such as mapping, monitoring, and target tracking. Such tasks often take the form of hard combinatorial problems. Hence, current algorithms are hindered from real-world deployment due to high runtimes caused by high amounts of inter-agent communication and lengthy communication messages, and due to unreliability caused by a lack of robustness to incorrect predictions. I will first present a coordination algorithm that enables self-configurable networks that balance resource reservation vs. coordination performance via balancing centralized vs. decentralized coordination. To this end, I will quantify the suboptimality cost due to decentralization by introducing a notion of mutual entropy for distributed submodular settings. I will then present online learning capabilities that enable the algorithm to operate in unpredictable environments, and even enhance its performance by robustly leveraging untrustworthy external commands such as commands from human operators. To this end, I will present tools for bandit submodular optimization, generalizing to the multi-agent setting expert algorithms from operations research for single-agent decision making. I evaluate all algorithms in area monitoring and target tracking scenarios in AirSim, simulating communication delays. I will conclude with complementary results and ongoing research.

Speaker's Bio

Vasileios is an Assistant Professor in the Department of Aerospace Engineering, at the University of Michigan, Ann Arbor. Previously, he was at the Massachusetts Institute of Technology (MIT), in the Department of Aeronautics and Astronautics, and in the Laboratory for Information and Decision Systems (LIDS), where he was a research scientist (2019-2020), and a post-doctoral associate (2018-2019). Vasileios received his Ph.D. in Electrical and Systems Engineering at the University of Pennsylvania (2018). Vasileios cares for scalable and reliable cyber-physical systems in resource-constrained, unstructured, and contested environments, such as the systems found in defense, disaster response, and smart cities. He thus works at the intersection of control, robotics, and online learning, contributing, on the cyber side, resource-aware optimization algorithms for perception, control, and multi-agent coordination, and, on the physical side, morphable aerial vehicles for superior control authority. Vasileios is a recipient of the Best Paper Award in Robot Vision at the 2020 IEEE International Conference on Robotics and Automation (ICRA), of an Honorable Mention from the 2020 IEEE Robotics and Automation Letters (RA-L), and was a Best Student Paper Award finalist at the 2017 IEEE Conference in Decision and Control (CDC).

Video URL: