CONTROL AND TRAINING OF ROBOTS USING LARGE LANGUAGE MODELS

Smart overview of the Invention

The patent application describes a method for controlling and training robots, specifically collaborative robots (cobots), using large language models (LLMs). The process involves sending a query to either a cobot or an LLM to trigger the cobot to perform a specific task. The LLM is constrained to provide responses in the form of human-readable discrete tasks that correspond to computer-readable code, which directs the cobot to achieve the target objective. The method includes auditing these tasks before authorizing the cobot to execute the code.

Field of Invention

The invention pertains to the integration of LLMs with robotic systems for improved control and training. LLMs, like ChatGPT, use deep learning algorithms to interpret and generate text based on extensive datasets. This technology is leveraged to enhance the interaction between humans and cobots by translating natural language queries into actionable tasks for robots.

Background and Context

LLMs can sometimes produce unexpected or inappropriate outputs, which can be problematic when generating code for robots. Cobots are designed to work alongside humans in shared environments, making safety a critical concern. The invention addresses potential risks by ensuring that tasks generated by LLMs are auditable and safe before implementation. Cobots can be used across various sectors, including residential, industrial, and medical applications.

Implementation Details

The method involves using a local computing device for user interaction with both the LLM and the cobot. Users can issue commands in natural language, which are then processed by the LLM to generate computer-readable code for the cobot. The system allows users to set parameters for task generation, ensuring tasks are discrete and auditable. The cobot can store known locations and use sensors to map environments, enhancing its ability to execute tasks accurately.

Practical Applications

Cobots equipped with this system can perform navigation and environmental interaction tasks efficiently. Navigation tasks might include precise movements like "move forward five feet" or "turn left ninety degrees." Environmental tasks could involve actions such as "pick up" or "deliver" objects. By converting these commands into base-level functions, cobots can safely interact with their surroundings, executing tasks through computer-readable code generated by the LLM based on human-readable instructions.