A new study from MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL) demonstrates that large language models (LLMs) can be used to generate detailed, step-by-step robotic task plans directly from natural language instructions, bypassing the need for traditional, labor-intensive programming. The system, named ‘Code as Policies,’ allows robots to interpret high-level commands, such as ‘set the table,’ and autonomously generate the necessary code to perform complex sequences of actions involving multiple objects and tools. Researchers tested the approach on real robotic arms in simulated kitchen and lab environments, where the models successfully created plans for tasks like arranging blocks and using tools. The work highlights a shift toward more accessible and flexible robot programming but notes current limitations in handling physical errors or highly unpredictable environments. For the full details, read the complete article at the source.