MIT robotics researchers invented a new way to make robots smarter. They call it Relevance. This method helps robots focus on important things. Robots often struggle with too much information. They see and hear everything in a scene. Deciding how to help humans takes a lot of effort. Relevance uses audio and visual cues. The robot guesses what a human wants. Then it finds objects to help with that goal.
The researchers tested this at a pretend breakfast buffet. They set up a table with fruits, drinks, and snacks. A robotic arm with a camera and microphone helped out. In one test, a person reached for a coffee can. The robot saw this and gave milk and a stir stick. In another test, two people talked about coffee. The robot heard this and brought coffee and creamer. The robot predicted human goals correctly 90 percent of the time. It picked the right objects 96 percent of the time. Relevance also made the robot safer. It reduced collisions by over 60 percent compared to older methods.
How Relevance Works
Relevance mimics how human brains focus. Humans use the Reticular Activating System, or RAS. RAS is a brain part that filters out unimportant things. It helps people focus on what matters, like making coffee. The robot system copies this idea. It has four steps. First, the robot watches and learns. It uses a microphone and camera to collect information. An artificial intelligence (AI) toolkit processes this data. The AI toolkit includes a large language model (LLM) and various algorithms.
The robot checks if a human is nearby. If it finds one, it focuses on helping. It uses the AI toolkit to guess the human’s goal. For example, it hears “coffee” and sees a reaching hand. It decides the goal is making coffee. Then it picks relevant objects like cups, not snacks. Finally, the robot plans a safe path to deliver the objects. Researchers want to use this in workplaces and homes next. They hope robots can help with tasks like laundry or repairs.