Hi, I’m Benjamin Drumwright, an engineering student focused on digital design, FPGA/ASIC development, and embedded systems. This website is where I document the work that reflects who I am as an engineer. I built this space as a living portfolio to showcase my technical abilities, problem-solving mindset, and commitment to learning. Every project featured here represents a hands-on application of the concepts I’m passionate about and a step toward becoming a well-rounded engineer.

This project is a modern C++ recreation of the classic Oregon Trail game, where players take on the role of a trail leader guiding settlers across the American frontier. Users choose their profession, name their party, manage resources, and make critical decisions at settlements, rivers, and random events. Key game mechanics include resource tracking, random event generation, player choices, and dynamic settlements with changing prices.

The code implements game loops, condition checks, class interactions, and structured randomness to simulate challenges such as sickness, mechanical failures, and river crossings. It also features dynamic scoring based on survival and inventory management.

The goal of this project was to create an autonomous item delivery robot using the Boston Dynamics Spot robot. This robot was intended to identify and deliver specific items to designated locations using computer vision and a flexible, programmable control system. The project combined robotics, embedded systems, machine learning, and software development into a fully functional solution capable of real-time item recognition and delivery.

To accomplish this, we used Boston Dynamics’ Spot robot as the mobile base platform, leveraging its advanced mobility and built-in sensors. The computational backbone of the system was an NVIDIA Jetson, which processed visual data using a custom-trained TensorFlow image recognition model. This allowed the robot to recognize specific objects or packages and determine the correct delivery path based on visual input. We also made extensive use of Spot’s built-in SDK and Mission Control system to program and execute delivery tasks.

For this project, I designed and implemented a custom neural network entirely in VHDL to perform image classification of fruits (apple, banana, orange) using a PYNQ-Z2 FPGA and a Raspberry Pi 5. The goal was to understand neural network hardware implementation from the ground up without relying on prebuilt accelerators or high-level frameworks.

Project Overview

The system was divided into three parts:

Model Training (TensorFlow)

A lightweight CNN was trained using TensorFlow on grayscale 32×32 image patches of fruits. The model included a convolutional layer, ReLU activation, and a fully connected output layer. After training, weights and biases were extracted and converted into fixed-point representations suitable for hardware.