Charles Benjamin

I helped create PEARL 3.5, MIT’s robotic boat and mobile docking platform. Alongside the boat’s autonomy and field operations, I led aircraft integration through a river landing and built coordination software for flying missions, finding PEARL, and returning to it.

I prepared boats and missions, helped teach an MIT marine autonomy course in Piraeus, and tested an AI teaching assistant built with Google’s NotebookLM. The experience culminated in a presentation to Prime Minister Kyriakos Mitsotakis, Defense Minister Nikos Dendias, and senior Greek military officials.

In Norway, I worked on RAN, a former Formula 2 racing boat converted into an uncrewed vessel. I developed its autonomy, helped operate it, and adapted a radio-controlled plane for autonomous flight alongside it. I also guest-judged the Njord Autonomous Ship Challenge, an international student competition.

I built connected workflows for AI coding assistants using MOOS-IvP, an open-source framework for robot autonomy. The plugin guides them through developing apps, missions, and tests. I then benchmarked how that guidance affects the work they deliver.

I built a continuous integration and delivery (CI/CD) pipeline for MOOS-IvP’s robot-autonomy software. It automates code checks, simulated mission tests, and macOS and Linux builds. Core developers have used it to test proposed changes, catching bugs in timing, coordinate conversion, and viewer configuration.

My aircraft work in Norway inspired a year-long McGill project: documenting and testing a fixed-wing drone platform that uses MOOS-IvP for mission decisions and ArduPilot for flight control. I also simulated aircraft bidding to replace one another as their range ran low. Parts now inform related master’s research at MIT.

I’m interested in how marine robots divide up work when communication is unreliable. I implemented the Consensus-Based Bundle Algorithm (CBBA), which lets vehicles bid for tasks without a central planner, and related methods in MOOS-IvP to study how teams agree on a plan, carry it out, and confirm it’s complete.

I built moos-map to make reliable map backgrounds for robot missions, extending earlier community tools with exact geographic cropping, viewer-compatible files, and automated output checks. It works through a browser interface or command line, and I published it on PyPI, the Python package index, for others to install and use.

Helping friends with MIT’s marine autonomy course, 2.680, inspired me to expand the community’s extension for Visual Studio Code, a code editor. I added parameter definitions on hover, formatting, and configuration checks, including warnings for non-convex mission regions: polygons with inward corners that certain vehicle behaviors can’t use.

I wanted to share MOOS-IvP runs without recording my desktop. I built and published alog2media to turn mission logs into videos, GIFs, and still images, making experiments easier to document, present, and inspect.

I’m studying MAVROS, a bridge between drone autopilots and the Robot Operating System (ROS) framework, to design similar services for sending commands and receiving flight data through MOOS, MOOS-IvP’s messaging layer. The motivation comes from PEARL, Norway, and my fixed-wing aircraft project. Implementation is planned for this semester.