syracuselocal.org

Cornell's RHyME Enables Robots to Learn from Videos

Cornell University researchers have developed a new robotic framework called RHyME—short for Retrieval for Hybrid Imitation under Mismatched Execution—that allows robots to learn tasks by watching a single how-to video. The system uses artificial intelligence to translate human demonstrations into actions that robots can follow. By reducing the need for extensive, step-by-step task instructions, the framework could decrease the training time, energy, and money needed to deploy robotic systems.

The research team, led by doctoral student Kushal Kedia and assistant professor Sanjiban Choudhury, is using what is known as imitation learning. This approach enables robots to learn a sequence of tasks more quickly and helps the machines adapt to real-world environments. The team compared the AI translation process to converting between languages, saying it is similar to translating French to English by converting human demonstrations into robotic actions.

The development comes as part of efforts to improve home robot assistants and other robotic applications, addressing challenges faced when robots lose tools or screws during tasks. Further details on the framework and its implementation are available through Cornell University's publications.

Source: Cornell University

Source Type: College

Source Location: Ithaca

Apr 23, 2025 at 8:30 AM

College