A tiny autonomous Pac-Man machine with one job: learn the maze, chase the points, and try not to bonk into anything important.
PacBot spent most of the year building a new iteration of the robot. The design is more compact, easier to repair in pieces, and packed with better components. Last year's competition exposed the weak spots, so the team took notes and got busy.
The rebuild paid off at the 2026 PacBot Competition at RIT, where Columbia placed SECOND!
One little robot still requires plenty of work. The team splits that work into three subgroups, each with its own way of making PacBot faster, smarter, and less likely to need an emergency fix.
Design the robot hardware, build the maze, and track down components that fit without playing enclosure Tetris.
Design PCBs, solder everything together, and keep the sensors talking to the rest of the robot.
Work on reinforcement learning, A* pathfinding, embedded programming, and PID control.
The goal is simple on paper: score as many points as possible in a physical Pac-Man maze. The hardware has to be compact, fast, and maneuverable, while carrying the circuits, motors, and sensors that make autonomous driving possible.
The high-level software finds a good route through the maze in real time. Down below, embedded code reads sensor data, controls the motors, and keeps the robot stable and pointed where it thinks it is pointed. That last part is sneakier than it sounds.
