I’m just a kid building stuff I find cool — embedded systems, robotics, applied AI, and much, much more. I focus on practical end-to-end projects: hardware design and firmware, control and estimation, software tooling, and AI and ML models that are deployable in real systems.
Contact me on shreeshtripathi13@gmail.com for more info and collaboration options.
-111827?style=for-the-badge&logo=onshape&logoColor=22D3EE){kind=icon}
| Area | Topics & Focus of current learning and projects |
|---|---|
| Robotics & Embedded Systems | STM32 firmware & peripherals, real-time RTOS workflows (FreeRTOS), sensor fusion & state estimation (IMU fusion, Kalman filters, complementary filters), odometry & localization, motor control and feedback loops, hardware debugging & instrumentation, mechanical-electrical co-design |
| Control Theory & Estimation | Classical control (PID tuning, frequency-domain analysis), state-space methods, LQR/LQG, model predictive control (MPC), observer design, Kalman filtering and smoothers, system identification, robustness and stability margins |
| Agentic Systems & Language-Integrated Agents | Reinforcement learning fundamentals (policy/value methods, off-policy learning), planning + perception loops, LangChain-style agent pipelines, tool use by language agents, language graphs and knowledge-graph grounding for agent memory and reasoning |
| AI / ML / Computer Vision (Concrete & Applied) | Probabilistic modeling & optimization, representation learning, convolutional and transformer architectures, PyTorch internals (autograd, custom ops, efficient dataloaders), training at scale (mixed precision, distributed training), practical CV pipelines for robotics (detection, tracking, depth estimation) |
| Software & Systems Practice | C/C++ systems programming, embedded debugging patterns, performance profiling, tooling and reproducible experiments, data pipelines and evaluation metrics |
| CAD & Product Thinking | Parametric design, tolerance-aware assemblies, iteration for manufacturability, integration of mechanical design with sensing and actuation |
Also a big chunk of what I have currently learned, I haven't pushed actual repos for those tiny, learning-based projects, I'll be commiting more during my next vacation as I have just entered high school.
This is my flagship project that brings together most of my areas of focus.
-
Perception: onboard computer vision for object detection, semantic segmentation, and optical flow; depth estimation and sensor fusion (camera + IMU) for robust situational awareness (using OpenCV).
-
Navigation & Autonomy: ROS 2-based perception-to-planning stack, SLAM / visual-inertial odometry, waypoint planning, and reactive obstacle avoidance; integration of higher-level planning (mission scripts, behavior trees) with low-level controllers.
-
Control & Estimation: state estimation (Kalman/EKF), PID / LQR controllers for attitude/position, and experiments with MPC for constrained trajectory tracking.
-
Embedded & Firmware: STM32 flight controller module (Inbuilt STM32 on Uno Q), Zephyr RTOS-based, telemetry and fail-safe behaviors, real-time logging for post-flight analysis.
-
Mechanical & CAD: Onshape-driven iterative airframe design, vibration isolation, sensor mounting strategies, and manufacturability considerations.
-
ML & Deployment: Custom YOLO model, TorchScript/ONNX deployment, and runtime optimizations for onboard compute (specifically for the Arduino Uno Q).
-
Pixhawk/PX4 Integration: Onboard Pixhawk 6C (maybe, if flight alogorithm is not on Uno Q itself). MAVLink and MAVSDK integration possible.
-
Simulation first focus in Gazebo, with optimized ROS component and ROS Control integration.
The interceptor drone is a work-in-progress — repositories above are used as building blocks. I regularly push firmware updates, perception experiments, and integration tests; see the repo list for links.
I am also currently working on an intelligent robotic arm that combines the following key components and goals:
- Also a focus on simulation.
- Using the Arduino Uno Q as the brain.
- Running a lightweight VLA model on the cloud.
- ROS2 and ros2_control integration.
- NLP control.
- Inverse kinematics built from scratch.
- Inexpensive.
- Fully open source and documented.