Rocky's Open-Source Build Thread (AI for Industry Challenge)

Rocky Shao, a freshman at The Ohio State University with limited robotics experience, is publicly documenting his entry for the AI for Industry Challenge—an open-source project to create an AI system capable of performing industrial cable insertion tasks. His GitHub repository shows he's building on top of the LeRobot framework, specifically modifying the `lerobotTeleop` code to create a 'Data Collection' class. His technical plan involves generating synthetic training data using ground-truth transformation frames (a method for precise robot positioning), then using $300 in student credits to train a LeRobot ACT model via Google Colab. The goal is to move beyond hard-coded movements to a policy that can stop based on real force feedback.

Shao's current progress, posted on March 12, 2026, includes a working 'CheatCode' teleop mode that successfully automates cable insertion for the first two challenge trials. He's troubleshooting a naming mismatch for the third trial's plug type. His hardware setup is a consumer-grade Asus Zephyrus G16 laptop running Ubuntu 24.04, proving the project's accessibility. The next steps involve learning to correctly record multiple training episodes in LeRobot and pushing the collected dataset to Hugging Face. His thread highlights the growing trend of students using open-source robotics stacks like LeRobot and ROS 2 to tackle complex manipulation problems previously reserved for well-funded labs.