Master's-thesis control engineer at FAU Erlangen-Nürnberg, based in Bavaria. Writing my thesis on SLQP-MPC for a Quanser RT2 linear inverted pendulum. Looking for a Werkstudent position in control, estimation, signal processing, robotics or embedded systems.

📍 Nürnberg, Germany · 📧 sreeramanil30@gmail.com

A header-only C++17 library and benchmark of multi purpose 86 state estimators and observers. Implemented for battery-management systems in electric aircraft, with aerospace attitude-estimation heritage. Every algorithm is an original implementation from its primary publication, embedded-safe by construction (no heap, no exceptions, float32-ready, Cortex-M7 verified in CI), and cross-validated against FilterPy, PyBaMM and ahrs.

Repo: estkit · Site: anilram30.github.io/estkit · Report: 958 pages, one chapter per estimator

86 estimators · 12 families · 2 truth plants · 12 fault scenarios · ~9,000 benchmark runs · 0 runtime dependencies

Code Apache 2.0, report and data CC BY 4.0.

Site: anilram30.github.io/Apollo-Trajectory-Recreation

The first Kalman filter to fly — NASA TR R-135 (Smith, Schmidt & McGee, 1962) — reconstructed in MATLAB from the original documents and flown over a complete Earth → Moon → Earth ballistic free-return mission. A genuine figure-8 trajectory targeted in the full Earth(J2) + Moon + Sun field, optical-angle sightings with the horizon-altitude bias measured on Apollo 13's own P23 data, and both variants of the filter: the nominal-linearized form as first published, and the estimate-linearized form R-135 already recommended and history later named the EKF.

Repo: Apollo-Trajectory-Recreation · Report: 13-page narrative through the primary NASA sources .

5.72-day free-return mission · 10-state augmented filter · 1,986 sextant sightings · 31,000 km open-loop miss → 1.4 km EKF error · 2 MATLAB files

Code Apache 2.0, report and data CC BY 4.0.

Distributed model-predictive control and moving-horizon estimation for a heterogeneous UAV cluster — quaternion hexacopters and 6-DOF fixed-wing aircraft, each agent running its own constrained real-time estimator and controller, coordinating over a communication graph with no central node. Six hexacopters hold formation through a ten-second GNSS blackout: the distributed estimator fuses relative measurements to GNSS-good neighbours and pins each denied agent to within nine centimetres of truth while dead reckoning drifts past a metre. The anchor-coverage condition on the communication graph is identified as the structural limit, and a leveled multi-hop anchoring extension recovers the broken case with the safety margin restored. The report derives both airframes from first principles; the compiled real-time solvers substituted into the closed loop reproduce the reference behaviour without change.

Repo: uav-cluster-dmpc-dmhe · Site: anilram30.github.io/uav-cluster-dmpc-dmhe · 3D replay: SwarmScope · Report: 57 pages, first-principles derivations of both airframes

6 hexacopters + 4 fixed-wing aircraft · 13-state quaternion plants · 210-run Monte-Carlo campaign, 0 divergences · dead-reckoning 1.26 m → DMHE 0.088 m · min separation 1.55 m against 1.4 m barrier · compiled solvers in-loop ≈76 µs / agent / cycle

Code Apache 2.0, report and data CC BY 4.0.

Output-feedback trotting control of a 45 kg ANYmal-C in MuJoCo, using a nonlinear model-predictive controller and nonlinear moving-horizon estimator that share one single-rigid-body model, one symbolic source and one design language. The estimator reconstructs the base state and an external disturbance wrench from proprioception only — IMU, leg kinematics and joint torques — while the controller plans ground reaction forces inside friction cones over a half-second horizon at 100 Hz and feeds the estimated disturbance forward. The robot ramps to a trot, absorbs a 40 N lateral push and turns, with no ground-truth state anywhere in the loop. The report derives the rigid-body model, gait, footholds and measurement model from first principles; the compiled real-time solvers substituted into the closed loop reproduce the reference behaviour without change.

Repo: ANYmal-C framework · Site: anilram30.github.io/anymal-srbd-mpc-mhe ·

45 kg ANYmal-C · 18-DoF MuJoCo plant · 12-state SRBD · 18-state NMHE with 6-state disturbance wrench · NMPC N=50 at 100 Hz · NMHE M=20 at 100 Hz · 40 N push estimated at 39.5 N · lateral drift 0.15 m vs 0.74 m without force feed-forward · 5/5 noise seeds · stable to 2× nominal sensor noise · NMPC 56 µs feedback · NMHE 37 µs feedback

Code Apache 2.0; report and results are included in the repository.

Seven engineering packages that take a high-frequency cable from raw network-analyser measurement to a predicted automotive-Ethernet link. Each package has its own test suite, command-line interface, technical report and CI.

Site: anilram30.github.io/hf-cable · Hub: hf-cable-toolchain

• Real-time-iteration NMPC and NMHE solvers — ultra-fast solvers with state and input constraints, targeted at embedded control. (completed, will not be published, pending research)

• AEROFORGE — Autonomous aerial construction — heterogeneous UAVs cooperatively assemble and verify a bridge using distributed estimation and real-time NMPC under cable, contact, wind, and actuator constraints. (completed, will not be published, pending research)

• Missile Guidance & Simulation — 3-DOF 3-DOF missile dynamics and guidance simulation covering trajectory propagation, aerodynamic effects, guidance laws, and interception scenarios.

• Missile Guidance & Simulation — 6-DOF Full 6-DOF rigid-body missile simulation using position, velocity, attitude/quaternion, and body-rate states, with aerodynamic forces/moments, thrust, gravity, control-surface effects, and guidance.

• Hexacopter Simulation & Control in a Mars Rover mission — ROS 2 Six-rotor UAV simulation implemented in ROS 2, covering vehicle dynamics, attitude/position control, simulation integration, and the foundation for future sensor-fusion and autonomous-flight work.

• Vehicle control basics lane assist using PID + MPC in python (Udemy-updating)

• Vehicle suspension control - Nonlinear system linearization, State-space and Laplace analysis, Stability and pole analysis, Modal analysis, MIMO control, Pole placement, Vehicle suspension controller design, PID + LQR + Resonance analysis, Advanced vehicle suspension control using PID, LQR, resonance analysis, tuning with AI, and dominant pole approximation. (Udemy-updating)

• A Part 23 / DAL C Autopilot: GNC Stack, Real-Time Realisation, Simulation and Benchmark

• Embedded validation — I am also validating selected control, estimation and robotics projects on Raspberry Pi, moving algorithms from simulation to real embedded hardware and measuring their computational performance and real-time behaviour.

Repositories will be published as each project reaches a shareable state/ after publication.

Languages C++17 · MATLAB · Python · C · some C# and Java Control and estimation MPC, MHE, Kalman and sigma-point filters, particle filters, moving-horizon estimation, quadratic programming, Monte Carlo, sensitivity analysis Hardware STM32 (Cortex-M7, STM32Cube, HAL), Quanser real-time systems Tooling git, CMake, pytest, ruff, GitHub Actions CI, pandoc + XeLaTeX for reports

Email sreeramanil30@gmail.com Location Nürnberg, Bavaria — open to relocation across Bavaria and Europe for the right role.

The engineering decisions, algorithm and system design, validation strategy, analysis and stated limitations are my own.