Princeton University researchers have used an artificial intelligence model developed by OpenAI (GPT-3) to train a robot named TidyBot to perform household tidying tasks.
While it’s very much a prototype at this stage, users can issue specific storage instructions for various objects, such as organising clothes in specific drawers or identifying items to be disposed of in the trash.
The research showcases the robot’s adaptability and its ability to infer generalised user preferences, applicable to future interactions. Leveraging the GPT-3 model, the TidyBot demonstrates a high accuracy rate of 91.2% when dealing with previously unseen objects in the benchmark dataset, indicating its potential for learning new objects with limited exposure.
Real-world tests using the TidyBot have also shown its capability to organise and store 85% of objects in various scenarios. This practical demonstration highlights its potential to assist with everyday tasks and simplify consumers’ lives.
The collaboration between Princeton University researchers, OpenAI’s GPT-3 model, and the TidyBot robot showcases advancements in robotics and language models, offering promising possibilities for the future.
ADD WORKGROUP