My research is primarily focused on distributed control, learning, and optimization, with applications to robotics and cyber-physical systems. I am particularly interested in developing novel theory and algorithms for autonomy of multi-robot systems under performance guarantees (e.g., safety, efficiency, robustness).
Sponsored Projects:
Context-Aware Reinforcement Learning with Complex Objectives and Constraints
Agency and Period: DARPA, 09/2021 - 02/2023
Description: The main objective of this proposal is to develop provably correct algorithms for reinforcement learning (RL) under complex objectives and constraints. Bounded temporal logics are used to express complex specifications. The project aims to develop a novel automata-theoretic approach for learning optimal policies while ensuring the satisfaction of temporal logic constraints in every episode of the learning process with a desired probability. Furthermore, the underlying risk-efficiency trade-off (i.e., how increasing the desired probability of constraint satisfaction reduces the expected reward) as well as the complexity/scalability of proposed algorithms will be investigated.
Space Vehicle Swarm Coordination and Control using Temporal Logic
Agency and Period: NASA, 05/2021 - 06/2022
Description: The main objective of this proposal is to develop and evaluate multi-agent control and planning methods for achieving complex swarm objectives. To this end, the project aims to first develop a novel temporal logic representation for encoding global objectives of possibly heterogeneous swarms. Second, the project aims to design distributed planning and control algorithms for agents to minimize their individual costs (e.g., fuel consumption) while collectively satisfying the global temporal logic specification.
Swarm-by-Logic: Swarm Autonomy with Receding Horizon Based Temporal Logic Framework
Agency and Period: DARPA, 09/2018 - 04/2019
Description: The overarching goal of this project is to develop a novel hierarchical architecture for swarm autonomy: the highest layer decomposes the swarm into smaller squads and generates the high-level plans for squads; the middle layer is responsible for decomposition of squad plans into individual objectives of robots expressed as signal temporal logics; and the lowest layer is responsible for control synthesis based on the temporal logic specifications. Flow of real-time information from lower layers to higher layers closes the feedback loop.