Robotics engineer. I build real systems that run on real hardware.
M.S. Robotics Engineering @ Wayne State University · Assistive & autonomous robots · Sim-to-real on quadrupeds
Most of what I build lives on a Unitree GO2 quadruped: perception, learned control, navigation, safety, and the unglamorous reliability layer that keeps it all from falling over in the real world. I like the part of robotics where the simulation lies to you and the hardware tells the truth. A lot of my repos document the bugs I hit getting there, because that is the part nobody writes down and the part that actually matters.
I work across the stack: C++ control loops, Python perception and RL, ROS 2 architecture, embedded firmware, CAD, and the GUIs on top. The list below is the stuff I am proud of, grouped by what it actually does.
An assistive quadruped that guides visually impaired users using audio-visual perception. The hard part is not following a person, it is following the right person and reacquiring them after they disappear behind an obstacle. The system covers audio localization, visual person re-identification, intent grounding, safety monitoring, custom ROS 2 messages, and a C++ gait controller, benchmarked against AprilTag-only, phone-only, and stock Unitree baselines.
ROS 2 Person Re-ID Whisper ASR GCC-PHAT Nav2 Jetson Orin NX Unitree GO2
A connected body of work turning a stock quadruped into something that learns, navigates, and recovers on its own.
| Project | What it does |
|---|---|
| go2-phoenix | Closed-loop sim-to-real locomotion learning in Isaac Lab, exported to ONNX and verified for train/deploy parity (max-abs diff 9.5e-7) before it ever touches the robot. |
| ashfall | Failure-driven reinforcement learning: the robot learns from where it falls, not just where it succeeds. |
| come-here | Hears "come here," localizes the voice, turns, finds the person, and walks to them. Audio-visual approach in one loop. |
| riskgraph-go2 | Persistent route-risk memory: the robot remembers where things went wrong and scores safer paths next time. |
| ros2-go2-nav2-yolo ⭐ | Full Gazebo autonomy stack (Nav2 + SLAM Toolbox + CHAMP + YOLOv8). YOLOv8n at 53 ms / 18.8 fps on CPU, with 10 non-obvious DDS/TF/SLAM bugs documented and fixed. |
The layer that tells you when your robot is about to do something stupid and catches it on the way down.
| Project | What it does |
|---|---|
| helix | ROS 2 fault-sensing: heartbeat monitoring, Z-score anomaly detection, log parsing, publishing structured FaultEvents instead of raw logs. Caught real LiDAR rate anomalies on a live GO2 graph. 96.5% TPR, 1.16 ms latency, 81K samples/sec. |
| BlackBoxRS | A flight recorder for ROS 2 robots: observability and post-failure forensics so you can actually answer "what happened" after a crash. |
| policy-health-monitor | Runtime safety net for learned policies: detects when a neural controller's internals go out of distribution and intervenes before it acts on garbage. |
| supercombo-blindspot ⭐ | Distribution-shift teardown of openpilot's production self-driving model. The question: does an L2 driving model know when it is blind? Threshold-free OOD metrics, recurrent-state-correct reproduction, the works. |
| Project | What it does |
|---|---|
| openvocab-tsdf | GPU-accelerated open-vocabulary 3D mapping: build a TSDF and query it in natural language. |
| go2-semantic-nav | Open-vocab 3D semantic scene graph feeding a language-grounded Nav2 overlay, running on Jetson Orin NX. |
Where I started, and still the most satisfying when it physically moves.
| Project | What it does |
|---|---|
| RADAR-Telepresence-Robot | Medical telepresence robot: remote teleop, live video with pan-tilt, and real-time SpO₂ / heart-rate monitoring through one Qt 6 operator console. |
| TicTacToe-3link-robot | A 3-DOF arm that computes closed-form IK to physically draw X's and O's while a Minimax AI plays optimally. ▶ Demo |
| EcoSort-bin | Multi-sensor fusion (weight + color + IR + ultrasonic) on an Arduino classifies and sorts waste. 97.5% accuracy over 40 trials, ~$70 BOM, zero ML. |
| Area | Technologies |
|---|---|
| Robotics | ROS 2, Nav2, SLAM Toolbox, ros2_control, tf2, Gazebo, Isaac Lab, MoveIt |
| Learning | PyTorch, reinforcement learning, sim-to-real, ONNX, out-of-distribution detection |
| Perception | YOLOv8, OpenCV, RealSense D435i, person re-ID, GCC-PHAT, open-vocab 3D mapping |
| Embedded | Jetson Orin NX, Arduino, Raspberry Pi, sensor/actuator integration, firmware |
| Languages | C++17, Python 3, MATLAB, C |
| Tools | Git, Docker, CMake, Qt 6, Fusion 360 |