I (We) worked on theoretical frameworks for RL algorithms with provably safe exploration AND convergence guarantees in collaboration with Dr. Wes Suttle, Prof. Ji Liu and Prof. Vijay Gupta. For simulations, I developed a safe quadrotor gym environment to illustrate our RL algorithms. At present, I am working to extend the current results (accepted in AISTATS 2024, code found here, paper here) under dynamical and environmental uncertainties.
I work on theoretical frameworks for safe controller design, using methods such as Control Barrier Functions (CBF) and Scenario Approach. At present, I'm working on non-convex CBF constraints as well as reducing the sample complexity in our earlier proposed risk-tunable safe controller, which made its way to IEEE CDC-2023 held at Singapore.
I worked on CBF based controller design for safe human-robot collaboration under dynamical and environmental uncertainties. This work is in collaboration with Prof. Yu She at his MARS lab. This work has been accepted in IEEE Transactions in Mechatronics with an option to present at AIM-2025 in Hangzhou, China.
I formulated a novel Markov Decision Process for the security constrained unit commitment problem, with states incorporating the inter-temporal constraints binding multiple horizons. Thence, I utilized TD learning algorithms to obtain the optimal generating schedules and dispatches.
I obtained novel mathematical results on Observability of LTI systems under unknown input and thence used these results to analyze the breach in privacy (plant states being the private information) achievable by an Eavesdropper, under the assumption that this eavesdropper has access to output data, input-update times and system matrices. The loss in plant-states' information is then illustrated with an example. This work led a publication in IEEE Control System Letters.
Part of the Open-Day demonstration, I designed and implemented a leader-follower coordination algorithm to make guard-robots maintain a uniform formation among themselves (by following the leader guard) and track the prisoner robot(often wirelessly controlled).
I implemented a multi-agent coordination algorithm, based on Kuramoto Oscillator, to align a set of robots uniformly on a ring. Thus obtained formations is also resilient to external disturbances.
Designing sampling instances such that the so-sampled LTI system becomes unobservable, I obtained mathematical result that showed that sampled-observability almost always comes back due to perturbations in the sampling times.
This is a course-project, where I considered unstable subsystems and studied the stability of the so- constructed switched system under minimum dwell-time switching signals. I used simulations to explore the stability of this systems with variations in minimum dwell time and found the limit for minimum dwell time such that stability is achieved.
This is a course-project, where I considered a mini power system model with hybrid state estimators(both PMU and SCADA measurements). Using sensitivity matrix, I divided the system into a set of islands with SCADA measurements. Thereafter, I used LAV(least absolute value) estimator to estimate the states of these individual islands at the rate of PMUs. I used MATLAB for the simulations.
This is a course-project, where I simulated the dynamics of the network states under 2.2 dynamical model for Synchronous generators, using the standard 10 bus-4 machine IEEE power system model. Thence, I created different types of faults in the system to characterize the stability under each fault and explore the corresponding minimum time the states took to diverge.
I used an LTI system to model guided missiles and used a model reference adaptive controller with MIT rule to control the pitch dynamics. I illustrated the performance of such controllers using simulations.