I am an R&D Engineer specializing in Embedded Control Systems and Physics-Informed Machine Learning. My work focuses on solving the "Reality Gap"—deploying intelligent agents onto hardware with strict latency constraints (<10ms).
I don't just train models; I deploy them to the metal.
| Domain | Technology | Use Case |
|---|---|---|
| Control & Dynamics | PID, LQR, System ID | Stabilizing unstable systems |
| Embedded Firmware | C++, PlatformIO, FreeRTOS | Hard real-time execution (Teensy 4.1, ESP32) |
| Scientific ML | PyTorch, JAX, ONNX | Physics-Informed Neural Networks (PINNs) |
| Simulation | Differentiable Physics | Sim-to-Real transfer & Domain Randomization |
Project: PIRL_Adaptive_Control
- The Challenge: Eliminating non-linear Stribeck friction in actuators without manual tuning.
- The Solution: A hybrid controller combining classical PID with a residual Physics-Informed Neural Network.
- The Constraint: Inference must happen in <500 microseconds on a microcontroller.
- Status: Phase 2 (Hardware Verification).
"We do not learn the robot; we learn the rust." Reliable control isn't about modeling the perfect physics; it's about robustly rejecting the imperfections (friction, noise, latency) that simulation misses.