Raaghav Thirumaligai ~ Controls portfolio

Control physical systems from models, data, and constraints.

I model dynamics, identify what matters, and turn feedback laws into behavior; from a 99-state distributed system to hardware that swings, balances, detects, and acts.

Explore the core projects Controls CV

99: state PDE model
Real time: Cart-pole hardware control
4: featured control + ML builds

Featured work · 01/04

From theory to practice

Each project follows evidence from model construction through closed-loop behavior, with visible tradeoffs and limits.

01 · Distributed-parameter systems · Apr-Jun 2026

Reduce the model.
Keep the behavior.

A boundary-actuated damped-wave PDE became a practical LQG study: discretize the field, shape disturbances, design output feedback, then ask how much controller order the full plant really needs.

Built a 99-state finite-difference plant with frequency-shaped disturbances and measurement noise.
Designed LQR and Kalman-filter controllers for three disturbance bandwidths.
Tested balanced-truncation orders 1-10 on the unreduced plant across damping regimes.

Transferable method: distributed-field modeling, reduction, disturbance estimation, and output feedback—the same toolkit used when state is spread across a thermal or energy system. This plant is a simulated linear fluid, not a thermal experiment.

LQGFinite differencesBalanced truncationMATLAB

Read the technical report ↗

Closed-loop frequency responses for full-order and reduced-order LQG controllers — Full-order response and reduced-order controller sweep on the same distributed plant.

99: plant states
1-10: reduced orders
3: noise bandwidths

Reduction exposed a real boundary: 29/30 tested controllers remained stable at highly damped β = 0.1, compared with 19/30 for the lightly damped β = 0.001 plant.

02 · Constrained optimization · Jan-Mar 2026

Trade speed for effort—on purpose.

Three controllers, one six-state orbital model, and the same randomized starts: a direct comparison of what changes when limits and fuel priorities enter the objective.

Zero-order-hold discretized Hill-Clohessy-Wiltshire relative dynamics at a 10-second sample time.
Implemented saturated LQR, state-focused constrained MPC, and fuel-prioritized economic MPC.
Compared feasibility, convergence, time, and cumulative L₁ input effort across 50 matched cases.

MPCEconomic MPCLQRMPCToolsMATLAB

Open the full project → Report ↗

Aggregate convergence, fuel proxy, and convergence-speed comparison for LQR and two MPC controllers — Fifty randomized rendezvous cases. “Fuel” is cumulative input effort, not propellant mass.

±0.5: mm/s² per axis
50: matched cases
3: controller priorities

03 · Physical control · Apr-Jun 2025

Identify it. Balance it. Catch it.

A physical cart-pendulum closed the gap between model identification and real-time behavior through estimation, energy shaping, local optimal control, and explicit mode transitions.

Collected logarithmic sine-sweep data and identified models around stable and unstable equilibria.
Combined energy-based swing-up with Kalman estimation and LQR/LQG balance and catch.
Ran the three-mode state machine on hardware through MATLAB, Simulink, and QUARC.

System IDLQR/LQGKalman filterHybrid controlQUARC

Open the full project → Hardware demo ↗

Swing-up, balancing, and catching control regions for the inverted pendulum — Three controllers coordinated by state-dependent switching. Rail impacts and encoder rollover remained explicit failure modes.

0.3-3: Hz identification sweep
30: experiment trials
3: control modes

04 · ML + embedded autonomy · Apr-Jun 2024

Perception driving motion.

C.L.E.A.N. brought classical machine learning into a physical loop: detect a can, steer toward it, verify the material, and coordinate a tracked base with a two-degree-of-freedom arm.

Led five people through weekly mechanical, electrical, and software build-test cycles.
Trained and tuned a nine-stage OpenCV cascade on approximately 1,500 custom images.
Integrated Raspberry Pi steering, Arduino motor control, ultrasonic and inductive sensing, and a worm-gear arm and claw.

OpenCVRaspberry PiArduinoState logicMechatronics

See the integrated prototype story →

C.L.E.A.N. project poster showing the autonomous robot, computer vision, mechanical system, electronics, and results — A ten-week team prototype. The classifier is a classical OpenCV cascade - not a neural network.

5: person team
~1,500: custom images
10: build-test weeks

Technical map

One practice, four kinds of evidence.

Move across the row to see where each project gets its model, how state is inferred, what closes the loop, and where the result was tested.

Methods across the four featured projects
Project	Model	Estimate	Control / decision	Evidence
Distributed PDE	Finite-difference acoustic field	Kalman filter	LQG + model reduction	Full-plant simulation
Satellite MPC	ZOH HCW dynamics	Full-state simulation	LQR + MPC + economic MPC	50 matched cases
Cart-pendulum	Experimental system ID	Kalman filter	Energy + LQR/LQG + modes	Real-time hardware
C.L.E.A.N.	Vision + sensor geometry	Cascade detector + sensors	Steering + state logic	Integrated robot

Research + experimental systems

Feedback thinking across biology, fluids, and instruments.

The common work is reconstructing dynamics from partial measurements, then building the model, detector, or apparatus needed to make the next decision trustworthy.

01Jan 2025 – Jan 2026

Biological Control Lab · Yeung UCSB

Transcription and DNA supercoiling as a coupled control system

Co-developed a control-oriented model of genome-wide transcription coupled to topoisomerase-regulated supercoiling. Used stability and sensitivity analysis on stochastic and nonlinear population models to identify parameters shaping transcription and growth.

Accepted manuscript · bioRxiv preprint ↗

02Aug 2024 - Jun 2025

M&M Flow Lab · Master's researcher

Detecting stochastic clogging with limited visibility

Studied particle-laden suspension clogging under changing material conditions. Prototyped a Python rolling-brightness and windowed-visibility detector for event logging and post-analysis; presented at APS DFD in November 2024.

PythonVision metricExperiment logsThermofluids

03Jan 2023 - Aug 2024

M&M Flow Lab · Undergraduate researcher

Instrumentation for repeatable granular-flow experiments

Designed and built an annular shear cell, then developed an Arduino-MATLAB strain-gauge calibration, filtering, and DAQ workflow. Presented the results at the Southern California Flow Physics Symposium.

ArduinoMATLABStrain gaugesFabrication

04Jun 2023 - Oct 2023

Fluid Energy Science Lab · Undergraduate researcher

Wind-tunnel measurement from apparatus to dataset

Collected wake-model verification data, built an Arduino-MATLAB DAQ system and app, and fabricated laser-cut and 3D-printed annuli for repeatable wind-energy experiments.

Wind tunnelDAQRapid fabricationModel validation

More analyses · supplementary work

More control ideas

Three compact studies extend the portfolio into hybrid biology, biologically inspired computation, and robust loop-shaping. Select a tab to open the evidence.

Modeled + simulated course research

Hybrid Dynamics of DNA Supercoiling Regulation

Modeled continuous mechanical relaxation interrupted by threshold- and dwell-time-triggered enzyme jumps, making a biological homeostasis question explicit as a hybrid automaton.

Defined flow and jump maps, enzyme-reset timing, and a homeostatic target band.
Derived a candidate dwell-time bound and explored weak forward invariance.
Ran MATLAB Hybrid Systems Toolbox ensembles and enzyme-action ablations.

Scope: derived and numerically explored; this version does not claim global asymptotic stability or strong forward invariance.

Open the report ↗

Hybrid automaton with DNA-supercoiling flow, topoisomerase jump, and gyrase jump modes — Flow and enzyme-triggered jump structure.

Simulated DNA supercoiling trajectory with continuous arcs, discrete jumps, and target band — One trajectory of DNA coiling over time exposes continuous arcs, jumps, and how evolution figured out a robust hysteresis control to maintain a level of stress.

Background

Control theory grounded in thermofluids and buildable systems.

Ongoing Ph.D. training in control theory, an M.S. focus in thermofluids, and four years of vehicle-system design anchor the work in physical dynamics and implementation.

Technical stack

Methods organized by what they do in the loop.

Control & Optimization: MPC, economic MPC, finite-horizon and constrained optimization, optimal control, LQR/LQG, nonlinear, robust, and hybrid control
Modeling & Estimation: State-space and distributed-parameter modeling, system identification, finite differences, balanced truncation, observers, Kalman filtering, model validation
Data & Machine Learning: Python, OpenCV, MATLAB, sensor and experiment-log analysis, classifier training and tuning, vision event detection and automation
Embedded & Real-Time: Simulink/QUARC, Raspberry Pi, Arduino, DAQ, sensor and actuator integration, state-machine logic
Physical Systems: Mechatronics, Formula SAE EV systems, instrumentation, design-build-test, SolidWorks, Fusion 360, fabrication

Education

UC Santa Barbara

Ph.D. Mechanical EngineeringControl Theory · GPA 3.92
Projected Jun 2029
Year 1 completed; available to take academic leave for full-time work.
M.S. Mechanical EngineeringThermofluids · GPA 3.90
Jul 2025
Experimental modeling, instrumentation, and data analysis.
B.S. Mechanical EngineeringProduct Design · GPA 3.76
Jun 2025
Capstone: 600 V Vehicle Battery Container ↗

Gaucho Racing Formula SAE EV · Aug 2021 - Jun 2025

Vehicle Systems Lead / Vice President

Four-year design-build-test contributor. Led a 600 V adhesive-joined accumulator container from CAD and drawings through structural checks; led braking-system design, manufacturing, testing, reliability, and a funded brake-fluid viscosity study.

UCSB Mechanical Engineering · Apr 2024 - Present

Teaching Assistant

Teach laboratory methods, dynamics, vibrations, and fluids through lab sections, office hours, assessment, and technical coaching.

Engineering Honors Program

Dean's Honors · 5 quarters

2 Undergraduate Research Awards · $3,750 + $2,500

Control physical systems from models, data, and constraints.