A new study from MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL) demonstrates that large language models (LLMs) can significantly accelerate the process of robot motion planning, a traditionally computationally heavy task. The research introduces a method where an LLM, such as GPT-4, breaks down a high-level user instruction into a sequence of manageable sub-tasks described in natural language. A separate, smaller AI model then translates each of these language-based steps into specific, actionable commands for the robot’s low-level controller. This hierarchical approach allows complex tasks like ‘place the apple in the pot’ to be executed more efficiently. In tests, the system dramatically reduced planning time and improved task success rates compared to previous methods, showing promise for more responsive and capable robots in dynamic environments like homes or warehouses. Read the full article at https://technologyreview.com/2024/07/12/1094755/llms-robot-motion-planning-mit/.