Cornell Combat Robotics · Autonomy subteam lead · in progress
Huey: Autonomous 3 lb Combat Robot
A beetleweight combat robot that finds and engages its opponent on its own — vision-based detection feeding an orientation controller, in an environment built to destroy it.
- weight class:
- 3 lb
Problem
Combat robotics is almost universally teleoperated — a human closes the loop. Huey's goal was to close it in software: detect the opponent, orient toward it, and drive the engagement autonomously, in a domain actively trying to break your sensors.
Constraints
- A 3 lb budget shared between armour, weapon, drivetrain, battery and the compute and camera for autonomy. Every gram spent on perception is a gram not spent surviving hits.
- Match-length decision latency: hesitation loses fights.
- Arena conditions are hostile to vision — harsh lighting, debris, motion blur, and an opponent whose appearance changes as it takes damage.
Approach
I led the autonomy subteam. A YOLO model detects the opponent from the onboard camera; an orientation PID controller with anti-windup turns detections into heading commands; closed-loop drivetrain control underneath keeps commanded motion matching actual motion as battery voltage sags and wheels lose grip.
Why it's technically hard
Everything is adversarial and non-stationary. The detector's subject is trying not to be where you predict, impacts knock sensors out of calibration mid-match, and the target leaves frame constantly — which is exactly why anti-windup matters, since a naive integrator saturates during every detection gap and then slams the robot past the target on reacquisition. All of it inside a weight and power budget a phone would find restrictive.
Status
Write-up in progress — competition results and the compute breakdown are still being collected.