Research agenda
Dependable, learning-enabled networked systems
I study cyber-physical systems whose success depends on more than a feasible schedule or a connected network. My research co-designs communication, computation, and control around three mission-level outcomes: estimation accuracy, timely data delivery, and closed-loop performance of learned policies.
Research thrust 01
Certified co-design of estimation, communication, and deployment
A network of agents can reach consensus and still agree on the wrong answer. I am developing a computable certificate for total estimation error, accounting for both disagreement among agents and error relative to the target. The goal is to use that certificate to choose communication links, allocate finite rates, and position mobile sensors around the quantity the mission actually needs.
For fixed teams, I am studying variance-weighted updates and a semi-distributed auction to choose links and rates. For mobile teams, I am studying how to couple network choices to positioning and use control barrier functions for safety and connectivity. I am also exploring a tracking-lag term for moving targets. This direction builds on my current estimation work and earlier methods for feasible mobile-sensor trajectories.
Research thrust 02
The aerial edge as a real-time platform
A UAV swarm is also a moving computing and communication platform. Where vehicles fly changes which links exist, how quickly data reaches a fusion center, and how much energy remains. This thrust targets timely delivery to the fusion center as formations and links change. I am investigating joint trajectory, routing, and rate assignment for UAVs and ground rovers relaying sensor data to fixed radios, alongside decisions about what to compute onboard or offload. I seek deadline guarantees that can be checked before flight and maintained through link loss and formation changes.
This thrust extends my doctoral research on timing guarantees for heterogeneous Time-Sensitive Networks to mobile wireless systems. It also builds on my work with the UC Irvine group validating distributed multi-UAV deployment on the AERPAW wireless testbed. Those experiments ground the analysis in onboard computation, vehicle control, and radio constraints.
Research thrust 03
Learning over networks that degrade what the learner sees
Distributed learning and control often assume that every agent can use an accurate, current global state. Real networks provide delayed, noisy, and sometimes corrupted information. My current work on reliability-gated state tracking examines how path reliability, channel noise, and the age of information should affect an agent's estimate, and when an unreliable update should be rejected. Here, the estimate is an input to learning and control, and the outcome of interest is the closed-loop performance of the learned policy.
I aim to quantify how network configuration affects learned-policy performance, study update gating as a sequential decision problem, and extend these ideas beyond linear-quadratic systems. The real-time question remains throughout: learned components have execution times, and the control loop has a deadline.
Research foundations
Guarantees tested against real systems
My earlier work asked whether a deadline-feasible schedule actually serves the control task. I optimized joint computation and network schedules for quality of control, developed ways to preserve capacity for future applications, and incorporated worst-case delay analysis for heterogeneous devices.
At UC Irvine, I extended this approach to mobile, wireless teams, where trajectory design must account for vehicle dynamics, energy, safety, and connectivity while still accomplishing sensing tasks. The three research thrusts carry the same principle forward: optimize the outcome that matters, prove what can be guaranteed, and check the result on hardware.