I build robots, autonomy systems, intelligent runtimes, and low-level engineering projects with an emphasis on measurable results, reproducibility, and explicit evidence.
Portfolio · Start with the robotic arm · Explore SLAM
Follow this profile for robotics builds, engineering experiments, benchmarks, and open-source systems.
|
Physical robotics Wearable MPU6050 gesture control over nRF24L01, driving a multi-joint robotic arm with real hardware actuation evidence. Stack: Arduino · C++ · MPU6050 · nRF24L01 · PCA9685 |
Autonomy & localization Reproducibility-first 2D LiDAR SLAM with Gazebo ground truth, ATE/RPE evaluation, loop-closure metrics, and rosbag regression. Stack: ROS 2 · Gazebo · LiDAR · Python · SLAM Toolbox |
Intelligent systems Local-first multi-model executive architecture with authority boundaries, durable missions, model routing, recovery, and verification. Stack: Python · local models · orchestration · security · systems architecture |
Physical Robotics
↓
Perception + SLAM
↓
Planning + Control
↓
Embedded Electronics
↓
Systems Engineering
↓
Intelligent Runtime Architecture
My repositories are meant to demonstrate different layers of the same engineering direction: building systems that sense, reason, act, recover, and can be measured honestly.
| Project | Engineering focus | Evidence status |
|---|---|---|
| gesture-controlled-robotic-arm | Wireless wearable control, sensor fusion, RF telemetry, servo actuation | Physical hardware demo and bench evidence published |
| slam-robot-ros2 | ROS 2 SLAM, Gazebo ground truth, ATE/RPE, loop closure, rosbag regression | Static contracts + ROS build CI verified; runtime benchmark evidence still gated |
| 3dof-robotic-arm | Analytic FK/IK, Cartesian planning, workspace/Jacobian analysis, Arduino control | Numerical/software validation in CI; physical accuracy evidence pending |
| custom-pcb-motor-driver | DRV8848 motor-driver PCB, current/thermal modeling, KiCad workflow | Analytical design evidence in CI; fabrication/bench evidence pending |
| http-server-from-scratch | C++20 HTTP/1.1 from raw sockets, secure static files, bounded concurrency, Linux epoll | Linux + Windows CI verified; controlled performance campaign pending |
| universal-brain | Local-first executive runtime, permissions, durable missions, multi-model routing, verification | Engineering checkpoints verified; target-machine endurance evidence still in progress |
- robotic-character-interface — safety-governed embodied AI, motion authorization, digital twin, telemetry, and fault testing.
- line-following-robot — control stack, PID behavior, regression testing, and robustness sweeps.
- cv-object-sorter — OpenCV perception → decision → actuation pipeline with evaluation tooling.
- gesture-controlled-robot — MediaPipe gesture control with temporal stabilization, STOP fail-safe, heartbeat, and watchdog behavior.
| Area | Technologies |
|---|---|
| Robotics | ROS 2, SLAM, Gazebo, TF, localization, kinematics, trajectory evaluation |
| Computer vision | OpenCV, MediaPipe, HSV/contour pipelines, offline evaluation |
| Embedded systems | Arduino, servo control, PCA9685, serial protocols, watchdogs |
| Electronics | KiCad, PCB design workflow, motor drivers, current/thermal modeling |
| Systems | C++20, raw sockets, HTTP/1.1, Linux epoll, concurrency, benchmarking |
| Software | Python, FastAPI, Flask, SQLite, React, TypeScript, automated testing |
| Engineering workflow | Git, GitHub Actions, CI, machine-readable evidence, reproducible runbooks |
I am currently pushing several projects from implemented to measured:
- SLAM: publish the first genuine Gazebo benchmark evidence bundle.
- 3-DOF arm: add real endpoint-accuracy and repeatability measurements.
- Motor-driver PCB: progress from analytical/CAD validation to fabrication and bench evidence.
- Universal Brain: execute real-environment Windows/WSL2/Ollama validation and endurance runs.
- Systems work: publish reproducible performance evidence for the C++ HTTP server.
This is where new results, failures, benchmarks, and lessons will appear.
Engineering principles
Simulation results stay simulation results. Analytical results stay analytical results. Physical claims require physical evidence.
Important experiments should preserve the exact commit, environment, configuration, raw artifacts, metrics, and failure cases so another developer can reproduce or challenge the result.
AI, perception, UI, and character layers should not directly command physical actuators. Motion authority belongs behind explicit planning, validation, safety supervision, and hardware boundaries.
Architecture, failure modes, limitations, evidence maturity, and release gates are treated as part of the engineering, not as afterthoughts.
If you work on robotics, ROS 2, SLAM, embedded control, computer vision, autonomous systems, or systems engineering, useful issues, experiments, benchmark reproductions, and technical discussions are welcome across the repositories.
Primary direction: robotics systems that are measurable, reproducible, safety-conscious, and explicit about what has actually been demonstrated.