As the development of artificial intelligence technologies continues to create revolutions in many fields, it also paves the way for significant innovations in the field of robotics. Researchers at the Massachusetts Institute of Technology (MIT) have announced that they are working on a new method that will allow robots to learn in a more flexible and agile manner. This innovative approach is inspired by the working principles of large language models, enabling robots to integrate information from different data sources and adapt more quickly.
Inspired by AI Models
In the past, robots were typically devices with limited capabilities, able to operate under specific conditions. However, MIT researchers have developed a model called ‘Heterogeneous Pre-Trained Transformers’ (HPT) to overcome these limitations. This model allows robots to process data from various cameras and sensors. Thus, robots can learn new tasks more quickly and effectively.
Researchers state that they were inspired by large language models like ChatGPT when developing this innovative method. The idea of large language models being pre-trained with a vast amount of language data and then becoming usable with minimal training provided a foundation for creating a similar learning process for robots.
The MIT team aims to further develop the potential of this method. Their ultimate goal is to create robot brains that can be downloaded and installed like an application. This means that robots will be able to perform a much wider variety of complex tasks. Researchers hope that these developments will usher in a new era in robot technology.