In an impressive demonstration of low-cost robotics and generative AI, two roboticists from UC Berkeley and ETH Zurich developed a $120 robot arm that can clean up spills, powered by GPT-4o. This project highlights how large language models are transforming robotics, even with budget-friendly components.
/
Building and Training the Robot Arm
Jannik Grothusen and Kaspar Janssen challenged themselves to create a functional robot arm using affordable materials. They successfully trained two $120 robot arms to respond to spills, relying on GPT-4o to guide the process. The team developed a visual language model specifically for human-robot interaction (HRI) within just four days, using around 100 demonstrations to train the robots’ movements.
Robot Arm DIY Plans Available
For those interested in building their own spill-cleaning robot, The Robot Studio has released a YouTube guide that walks through the process, making this practical robotics application accessible to DIY enthusiasts.
