Shayaan Shaikh

No earlier survey focused on learning for aerial manipulation, even as its use grows quickly. I analyzed the perception, control, and motion planning literature to map open challenges: object detection and localization with pretrained detectors fused with depth, system identification of unmanned aerial vehicle (UAV) and manipulator dynamics, and learning-augmented model predictive control and reinforcement learning.

Over 100 ST-16 star trackers already in orbit can watch for debris for free. I built a lightweight multi-object tracking pipeline in MATLAB (Kalman filters with Hungarian assignment) that separates resident space objects (RSOs) from background stars, reaching about 70% multi-object tracking accuracy (MOTA) on three flight datasets from Rocket Lab and SAIL, plus a confidence score that flags likely RSOs.

Modeled an underactuated 3-DoF Quanser helicopter from first principles and controlled it with cascaded PID loops in Simulink and on hardware, holding under 10% overshoot across its full range of motion. A Kalman filter fusing IMU and encoder data smoothed landings, a camera pipeline found drilling targets with color thresholding and blob detection, and drill control ran in C with limit-switch and IR feedback.

Led flight and ground software in Python and C/C++ for real-time telemetry (temperature, altitude, GPS) and a 2-DoF camera gimbal, prototyped on Arduino Uno and Teensy 4.1. The team received the TMU engineering faculty's Innovation Award in 2025.

Image: LeRobot

3D printed and assembled the open-source SO-101 arm and its teleoperation leader arm, then fine-tuned and deployed SmolVLA, a small open-source vision-language-action (VLA) model, with Hugging Face's LeRobot. Also trained a reinforcement learning policy in MuJoCo for lifting objects with the arm.

Fine-tuned a YOLOv11 detector (Ultralytics, PyTorch) to verify component placement on printed circuit boards (PCBs) across 13 component classes, reaching 81.7% mAP@50, alongside an OpenCV pipeline using image masking and edge detection.