Robotics
Jun 26, 2026
MIT researchers develop method for robots to understand vague instructions using AI
Jun 26, 2026
AI Summary
Researchers at MIT have created a system called Masked Inverse Reinforcement Learning (Masked IRL) that enables robots to interpret vague instructions and prioritize key details. This approach reduces the amount of demonstration data needed for training and improves robots' ability to navigate complex environments safely.
- MIT's Computer Science and Artificial Intelligence Laboratory (CSAIL) has developed a method called Masked Inverse Reinforcement Learning (Masked IRL) to enhance robot training.
- The system uses large language models (LLMs) to clarify ambiguous instructions and reduce the need for extensive demonstration data.
- Masked IRL allows robots to learn from kinesthetic demonstrations while capturing environmental information through sensors.
- The first LLM elaborates on unclear prompts, while a second LLM evaluates and prioritizes relevant details in the environment.
- This method has shown to improve robots' performance in tasks like navigating around obstacles and understanding user preferences without explicit instructions.
- In simulations, Masked IRL required fewer demonstrations to learn tasks compared to traditional methods and performed better when instructions were clarified.
- Future enhancements may include equipping robots with cameras to dynamically assess their surroundings and focus on relevant objects.
- The research was supported by the Tata Group and the Department of Defense, and findings will be presented at the 2026 IEEE International Conference on Robotics and Automation.
llmsrobotsinstructionsmitai