Arnav Arora

Contributed to a nationally competitive FTC team that has qualified for the state championship every season since my tenure began. Led rapid design, build, and test cycles using CAD-driven prototyping, followed by precision manufacturing via CNC machining and 3D printing. Designed and integrated compact, weight-optimized mechanisms including a reverse four-bar linkage, a differential 3-axis wrist, and a parallel-plate drivetrain. Implemented advanced control systems featuring object tracking, odometry-based localization, and finite state machines to improve autonomous accuracy and tele-operated efficiency, alongside fault-tolerant algorithms and positioning fail-safes. Continued to mentor and assist with new member skill- building in CAD, manufacturing, and programming.

controlmusic.art is a real-time hand gesture recognition system that enables touchless video playback control through machine learning. The application leverages MediaPipe Hands to detect and track 21 3D hand landmarks, extracting 63 features (x, y, z coordinates) that are normalized relative to the wrist position to achieve translation invariance. These normalized features feed into a supervised scikit-learn classifier trained on user-collected gesture data, predicting four control gestures: play, pause, rewind, and fast-forward. The Flask backend processes webcam frames sent via AJAX at 10 FPS. Client-side JavaScript captures video through the WebRTC getUserMedia API, renders frames to Canvas, and encodes them as base64 JPEGs for transmission. The server decodes frames with OpenCV, runs them through the MediaPipe-classifier pipeline, and returns gesture predictions and confidence scores as JSON responses that update the UI in real-time. The frontend features a responsive dual-pane layout with live webcam feed and real-time metrics, all styled with SCSS in a modern theme. The entire stack—MediaPipe, scikit-learn, Flask, OpenCV, and vanilla JavaScript—achieves approximately 100ms gesture recognition latency, demonstrating practical integration of computer vision and machine learning in an interactive web application.

Developed an ankle-based perturbation controller to enable realistic testing of lower-limb exoskeletal systems under trip and slip conditions. The system is inspired by Stanford Biomechatronics Lab's Bump'em framework, but redesigned for direct ankle perturbations rather than a torso-based system. Designed and integrated mechanical subsystems around existing lab hardware, including a Cubemars AKE60-8 motor and controller, and a metal mounting frame. Led CAD development and mechanical integration while contributing to closed-loop control design for scenario specific perturbations, improving experimental fidelity over prior capabilities which involved multidirectional impulses delivered to the user via impulses to the treadmill.