The Vending Machine Robot is a mobile snack-dispensing robot paired with a digital storefront, built from January to April 2026 by a multi-disciplinary team spanning mechanical engineering, electrical design, robotics software, and full-stack web development

Customers browse snacks online, pay through Stripe, receive a one-time code, and redeem it on the robot, which finds them with edge AI, opens its lid, and dispenses their purchase

Hardware: Raspberry Pi 4B, Intel NCS2, PiCamera Module 3, GPS, Pololu Maestro, dual H-bridge motor drivers, mecanum wheels with TPU rollers, and a welded 24-gauge steel chassis with 3D-printed PLA components

Software: Python, Flask, HTML, JavaScript, SQLite, Stripe, OpenVINO, and OpenCV

Libraries: threading, queue, socket, numpy, opencv-python, openvino, pygame, RPi.GPIO, pigpio, pyserial, bcrypt, requests, and python-dotenv

Mecanum wheels were selected so the robot could strafe and rotate in place, making it far easier to navigate crowded spaces and approach customers from any angle

Each wheel assembly combined 3D-printed PLA hubs and mounts with TPU rollers, bolted together and driven by dual H-bridge motor controllers on the Raspberry Pi

While PLA's relatively low strength meant some components experienced shearing under load, we addressed this with higher infill percentages, rounded geometries, and fillets to reduce stress concentrations in high-load areas

Welding thin-gauge steel with MIG meant every joint had to be executed with precise voltage, feed rate, and weld times as any discrepancy risked catastrophic burn-through

Each seam was planned carefully before striking an arc, and several practice runs on scrap were needed before the team was confident enough to weld the actual chassis panels

I designed the electrical system to accommodate devices requiring different voltages, manage heating concerns, and remain evolvable as parts arrived on staggered shipment schedules

The diagram mapped power distribution from the battery through buck converters to the Raspberry Pi, Pololu Maestro, motor drivers, NCS2, camera, and GPS, ensuring the AI team could begin software work even before every final component was installed

Mk-1 was the first physical layout of the electrical system as a proof-of-concept wiring pass that validated power routing and confirmed every device could be powered simultaneously without browning out

Lessons from this iteration informed cable management, connector placement, and which components needed dedicated regulators versus shared rails

Mk-3 is the production layout with every subsystem mounted, labeled, and routed for both reliability and serviceability inside the finished chassis

Battery limitations and heating from the NCS2 under sustained inference were addressed with dedicated cooling and a power budget that kept the Pi and AI chip stable through full purchase sequences

Behind a dedicated admin login, authorized users manage the robot's inventory, updating stock levels and pricing so the digital storefront always reflects what is physically available inside the machine

Every inventory change is logged to the backend actions table and forwarded to a Discord webhook for real-time audit visibility

The admin dashboard gives operators a single view of recent activity, heartbeat status, and action history, all clearly separated from the customer-facing experience

Admin authentication uses bcrypt for password verification, and Syan Shirazi performed penetration testing on both frontend and backend to ensure users cannot access critical information from the client side

To get a working system to the AI team with as much development time as possible, we built a cardboard robot that was structurally very similar to the desired end product

This stand-in let Samuel Saylor, Tri Nguyen, and Cayden Hutcheson begin tuning lid logic, screen integration, and customer-finding AI while the engineering team finished the real chassis

Over the course of printing, the 3D printer broke twice as the extruder badly jammed and the plate fell out of calibration, both requiring late-night disassembly to keep the project on schedule

Beyond hardware failures, classical printing problems piled up: adjusting infill percentages, print speeds, extruder and plate temperatures, brims, rafts, and orientations, and the time required to produce parts slowed iteration cycles enough that careful prioritization became essential

The full purchase sequence ties everything together: a customer completes payment on the website, receives a one-time code, enters it on the robot, and watches the lid open to dispense their snack

Low latency across the TCP bridge was critical; any delay between payment confirmation and robot response would break the seamless handoff from digital storefront to physical fulfillment

This was the highlight of college for me

Forget the long nights out at football games, forget Differential Equations, forget getting the occasional nervousness asking someone out, and forget those addictive, cream-stuffed cookies the campus restaurants sold to up-charge students

When I think about college, I will think about the pure, unadultered joy I felt working with the people I was so privileged to meet during my time at Rowdy Creators and specifically this project

Like all projects, there were bumps in the road, but for the rest of my life, I will be drawn to working with people simply because this (for the most part) went so well