My research interests are control theory, machine learning and optimization. I envision to develop intelligent control frameworks for decision making problems in complex dynamical systems.
Optimal multi-agent multi-target path planning in Markov decision processes
Missions for teams of autonomous systems often require agents to visit multiple targets in dynamic operating conditions where real time communication is difficult. Motivated by such scenarios, stochastic dynamics are used to model the motion of an agent as a Markov decision process. We showed that the single-agent problem is at least NP-complete by reducing it to a Hamiltonian path problem and proposed a polynomial time heuristic. For the multi-agent case, we proposed a heuristic partitioning procedure of assigning targets to agents that approximately minimizes the largest expected time to visit the target states. I presented a preliminary work at the 16th Coordinated Sciences Laboratory student conference which won the best poster award.
Explorative probabilistic planning with unknown target locations
Motivated by autonomous missions in remote or hostile areas, motion planning in an uncertain environment demands synthesis of an optimal control policy that balances between exploration and exploitation. In this work, we considered a generalized reach-avoid specification as the mission objective on a labeled graph environment and translate our problem to a Canadian traveler problem. We proposed a strategy by assigning edge weights that incrementally reveals the environment online while exploiting the current knowledge. I presented this work in 59th IEEE conference on Decision and Control and the presentation video is given below.