Research

Model-Based Reinforcement Learning for Resilient and Trustworthy Multi-Agent Decision Making

In this research thrust I establish theoretical foundations for coordination, control, and learning under uncertainty and adversarial influence, bridging theory and practice through real-world evaluations and unified performance metrics. My research focuses on routing and traffic management for large fleets of agents, developing adaptive algorithms with provable resilience and stability (defined as maintaining a uniformly bounded cost over time) guarantees. Ultimately, my research lays the groundwork for reliable autonomy in transportation, logistics, and mobility-on-demand services. Despite rapid progress in multi-robot coordination, navigation, and in reinforcement learning, real-world deployment remains challenging. Existing frameworks often assume fully cooperative or optimally performing agents, overlooking faults, uncertainty, or adversarial behaviors that can degrade performance or destabilize routing policies. My work addresses this gap by designing adversarially-aware planning and routing frameworks that maintain safety and performance even under adversarial interference or imperfect information.

I focus on resilient multi-agent decision-making under adversarial influence, developing model-based reinforcement-learning rollout routing algorithms that are provably robust to data manipulation, communication loss, or spoofing attacks, with results validated on large-city-scale real-world mobility data. These rollout methods offer theoretical guarantees for cost improvement, data efficiency, and online policy adaptation, making them well-suited for real-time control in safety-critical environments. My work specifically focuses on design of novel algorithms for large-scale multi-agent routing with provable time efficiency performance guarantees and for safe autonomous traffic management, resilient to faulty or adversarial agents, addressing regimes where existing theory fails. IEEE CDC 2025 IEEE CDC 2025 extended version

Multi-Agent Teamwork in Search for Smart Opponents Detection

In our works we have shown and quantified (for specific problems) gains obtained from carrying out a mission by a team of cooperating agents. The techniques developed can be applied in a variety of domains including planning against adversarial opponents, control of forest fires and search-and-rescue missions. A significant part of my Ph.D. work investigates team cooperation and planning under uncertainty in multi-agent search tasks, specifically for guaranteed detection of smart opponents. A smart opponent is an opponent that can perform optimal evasive maneuvers meant to avoid its interception. For these types of problems a "must-win" strategy is developed to ensure detection of all opponents in a region of interest. Our novel search methods and performance analyses are theoretically and experimentally proven for any number of searchers. Our State-of-the-art results on these subjects are available at: IEEE TRO 2021, Robotica 2021, J. Int. Rob. Sys. 2022, Frontiers on Robotics & AI 2023, ICR 2023, UR 2023, TAROS 2023, MRS 2023, Robotics and Autonomous Systems 2024, IEEE Transactions on Systems, Man and Cybernetics: Systems 2025

Urban Air Traffic Management of Autonomously Flying Vehicles

This part of my research concentrates on algorithms for safe and efficient multi-agent team cooperation under uncertainties. The study elaborates on a topic paramount for the advancement of large-scale traffic management algorithms for autonomous vehicles. We discuss urban air traffic management algorithms for futuristic autonomously flying vehicles by considering interactions between vehicle-like agent, where the term agent refers to an entity moving in the environment such as a vehicle. The investigated methodologies may be applicable to traffic management of autonomous ground vehicles as well. For more in depth details of my research on these topics see: RSS Pioneers 2023, RSS Towards Safe Autonomy Workshop 2024 For a preliminary version of an extensive survey we conducted on behalf of the Israeli Smart Transportation Research Center that investigates the current state-of-the-art in models and algorithms across many domains of intelligent transportation systems, see . ISTRC 2022 The findings were used to identify knowledge gaps in relevant areas of interest for further investigation.

Bio-Inspired Robotics

In this research we analyze locust trajectories in order to understand and model movements of locust swarms. The objective of this research is to assist in developing bio-inspired robotic technology by learning insect modes of locomotion towards bio-inspired locomotion. To this end, I developed a computer-vision and deep learning based multi-object video tracker, based on the principles of tracking by detection and association. It offers precise tracking for extremely long time horizons with state-of-the-art results on the problems of bio-inspired robots and insect tracking in dense environments. To achieve this goal, a new computer-vision software tool, called LocusTracker was developed for locust trajectory prediction. The tool enables tracking individual locusts in selected environments for very long time horizons of several hours without the usage of physical markers on individual locusts. For more in depth details of my research on these topics see: iScience 2024, Swarm 2024.