Our research lies broadly in safe and distributed robot autonomy.
Our research is primarily focused on developing scientific principles for enhancing the safety, robustness, efficiency, and versatility of autonomous robot systems. We are interested in designing autonomous systems that can perceive and reason about the ambient environment, adapt to unanticipated situations, and effectively coordinate with each other to accomplish complex, long-term, collaborative tasks. To realize this research, we develop new autonomy frameworks drawing tools from distributed control & optimization, multi-agent systems, task and motion planning, machine learning, active sensing and perception, and formal methods. A non-exhaustive list of our research projects can be found below.
Research Projects
Safe Perception-based Autonomy in Unstructured Semantic Environments [ARL CRA DCIST]
Recent advances in control theory, artificial intelligence (AI), and computer vision offer a tremendous opportunity to deploy mobile robots to uncharted, expansive, and dynamic environments to accomplish safety-critical missions. Key challenges involve quantifying uncertainty of AI-enabled components and their robust integration in the control loop. Towards this end, in this research project, we design novel autonomy architectures that tightly couple perception (e.g., cameras, LiDAR etc), AI-enabled object recognition methods (e.g., object detectors and classifiers) with novel planning and control methods. We aim to design autonomous systems that can reason about the geometric and semantic environmental structural, predict future environmental configurations, and make control decisions that adapt to online perceptual feedback.
Selected Publications:
Kantaros, Y., Kalluraya, S., Jin, Q., & Pappas, G. J. (2022). Perception-based temporal logic planning in uncertain semantic maps. IEEE Transactions on Robotics.
Kantaros, Y., Schlotfeldt, B., Atanasov, N., & Pappas, G. J. (2021). Sampling-based planning for non-myopic multi-robot information gathering. Autonomous Robots, 45(7), 1029-1046.
Resilient Multi-Robot Autonomy in Adversarial Environments
Several task and motion planning algorithms have been proposed recently to design paths for mobile robot teams with collaborative or conflicting high-level missions that are often specified using formal languages. A fundamental challenge in these works is to enhance robot resiliency against unanticipated, and possibly adversarial, events. In this research project, our goal is to design multi-robot systems that can react to (i) dynamic mission and safety requirements; (ii) unexpected changes in the structure of the environment; (iii) failures of collaborative robots; and (iv) adversarial behaviors of non-cooperative robots. Our approach relies on tools drawn from formal methods, motion planning, and adversarial and robust machine learning.
Selected Publications:
Vasilopoulos, V., Kantaros, Y., Pappas, G. J., & Koditschek, D. E. (2021). Reactive planning for mobile manipulation tasks in unexplored semantic environments. In IEEE International Conference on Robotics and Automation (ICRA).
Kalluraya, S., Pappas, G., Kantaros Y. (2023). Resilient Temporal Logic Planning in the Presence of Robot Failures
Trustworthy and Efficient Reinforcement Learning for Robot Control [NSF CPS]
Reinforcement learning (RL) has emerged as a prominent tool to control autonomous systems with highly nonlinear dynamics that are unknown or uncertain. Key challenges in RL include reward engineering, lack of safety/robustness guarantees, sample-inefficiency, and lack of generalization to unseen tasks. In this research project, by leveraging formal methods, we design trustworthy RL methods that can quickly learn verified control policies with provable sample-efficiency, performance, and safety guarantees.
Selected Publications:
Wang, Jun, Samarth Kalluraya, and Yiannis Kantaros (2022). “Verified Compositions of Neural Network Controllers for Temporal Logic Control Objectives.” IEEE Conference on Decision and Control (CDC).
Kantaros, Yiannis (2022) “Accelerated Reinforcement Learning for Temporal Logic Control Objectives.” IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS).
Distributed Autonomy in Communication-Denied Environments
Multi-robot systems are capable of accomplishing complex collaborative tasks ranging from transportation and delivery to surveillance and environmental monitoring. A fundamental requirement for efficient coordination is that of network connectivity. To meet this requirement, several controllers that ensure point-to-point or end-to-end network connectivity of mobile robot networks for all time have been proposed. Nevertheless, due to the uncertainty in the wireless channel, that affects signal strength in an unpredictable way, it is often impossible to ensure all-time connectivity in practice. Additionally, preserving all-time connectivity in communication-denied environments where communication is not readily available (e.g., underwater or underground environments) may significantly compromise mission performance as the robots are constantly subject to proximity constraints. In this research, we develop new distributed methods that enable intermittent connectivity in mobile robot teams operating in communication-restricted environments. While in disconnect mode, the robots can accomplish other tasks free of communication constraints.
Selected Publications
Kantaros, Y., Guo, M., & Zavlanos, M. M. (2019). Temporal logic task planning and intermittent connectivity control of mobile robot networks. IEEE Transactions on Automatic Control, 64(10), 4105-4120.
Khodayi-mehr, R., Kantaros, Y., & Zavlanos, M. M. (2019). Distributed state estimation using intermittently connected robot networks. IEEE Transactions on Robotics, 35(3), 709-724.
Formal Methods for Large-Scale Autonomy
Formal languages such as Linear Temporal Logic (LTL) have been extensively used in robotics due to their ability to capture high-level complex tasks, such as coverage, data gathering, persistent monitoring, and intermittent communication. Finding optimal robot paths that satisfy LTL-specified tasks can be achieved using tools from model checking theory and optimal control synthesis. The grand challenge in optimal control synthesis problems is dealing with the state-space explosion; in fact, existing approaches face scaling challenges as the number of robots, the complexity of the task, or the size of the environment increase. To mitigate these challenges, we have developed tools that can solve complex temporal logic planning problems for large-scale multi-robot systems (see e.g., our STyLuS* tool) with completeness, optimality, and fast convergence guarantees. Our tools have been applied in controlling mobile magnetic microrobots for manipulation tasks and building acoustic impedance maps using mobile robots.
Selected Publications
Kantaros, Yiannis, and Michael M. Zavlanos. “STyLuS*: A temporal logic optimal control synthesis algorithm for large-scale multi-robot systems.” The International Journal of Robotics Research 39.7 (2020): 812-836.
Kantaros, Y., Johnson, B. V., Chowdhury, S., Cappelleri, D. J., & Zavlanos, M. M. (2018). Control of magnetic microrobot teams for temporal micromanipulation tasks. IEEE Transactions on Robotics, 34(6), 1472-1489.