Team MacCody achieved best internal score of 151.71 but finished 65th (score 112.75) on the official leaderboard?

Team MacCody achieved best internal score of 151.71 but finished 65th (score 112.75) on the official leaderboard.

Used pi coding agent, GitHub Copilot, and a rented RTX 3090 to build a classical controller for UR5e in Gazebo?

Used pi coding agent, GitHub Copilot, and a rented RTX 3090 to build a classical controller for UR5e in Gazebo.

The approach was cheaper than training a policy but unable to beat top entries; the team calls it a mixed result.

Robotics

ROS Discourse May 27, 2026

⚡Two dads used LLM code generation to control a UR5e robot arm—here's what happened.

One email a day. Zero fluff. Join 10,000+ professionals.

Deep Dive

Key Points

Team MacCody achieved best internal score of 151.71 but finished 65th (score 112.75) on the official leaderboard.
Used pi coding agent, GitHub Copilot, and a rented RTX 3090 to build a classical controller for UR5e in Gazebo.
The approach was cheaper than training a policy but unable to beat top entries; the team calls it a mixed result.

Real-world test shows coding agents can speed up robotics prototyping, but classical controllers still lag behind learned policies in performance.