CONNECT WITH US

LLMs to pave way for direct human interaction to enter into home-use robot market, says DIGITIMES Research

Jin Pai, DIGITIMES Research, Taipei 0

Credit: AFP

The methods for humans to interact with and operate robots are gradually being simplified and the keen development of large language models (LLM) is expected to drive up the popularization of the natural language interaction method, which is intuitive and easy to work on, and DIGITIMES Research expects home-use robots to be the first to adopt the method.

Compared to professional AIs that are trained using specially prepared data and targeting specific tasks, LLMs can be applied to a variety of general missions. Microsoft and Alphabet have both developed interfaces for their interactive robots with LLMs that accept natural language commands and transfer them into planned movements.

So far, the operation of robots is focused on offline simulation, teaching pendants, or manual guidance, while the use of the natural language interactive method is also limited to specific missions with restricted functionalities.

Microsoft invested in OpenAI to develop Generative Pre-trained Transformer-3 (GPT-3) model and the ChatGPT platform, while UK-based Engineered Arts' humanoid robot Ameca has adopted GPT-3 to make conversation with humans.

Microsoft's ChatGPT robot's conversation interface is able to accept natural language commands that can output a JavaScript Object Notation (JSON) format movement plan.

Alphabet's PaLM-SayCan based on Pathways Language Model (PaLM) can accept natural language commands with unclear semantics and can analyze them into workable movement plans. PaLM-E is a system that combines the PaLM language model with a visual model and can understand the environment with visual data and output movement plan via natural language commands. The system is able to resolve the drawback that a language model can only interact with human reality via text instead of direct interactions.

Although the market has not seen any availability of robots with the support of LLM, the potential of LLM is still enormous. Home-use robots, which do not require high movement precision, are expected to adopt LLM at a much faster pace than industrial robots, DIGITIMES Research expects.