{
  "id": 19096823,
  "title": "AI-assisted learning platform helps robots reason beyond literal commands",
  "url": "https://urgent.news/2026/11/09/ai-assisted-learning-platform-helps-robots-reason-beyond-literal",
  "topic": "ai",
  "section": "AI",
  "published": "2026-11-09T05:00:00.000Z",
  "source": {
    "name": "MIT News Research",
    "slug": "mit-news-research",
    "url": "https://news.mit.edu/2026/ai-assisted-learning-platform-helps-robots-reason-beyond-literal-commands-1109"
  },
  "original_language": "en",
  "account": "A new artificial intelligence-assisted learning platform called FLEX, developed by MIT engineers, is giving robots the ability to understand human language commands beyond their literal meaning. FLEX, or Flexible Feature-Editing Pipeline for Robot Motion-Planning, allows humans to interact with robots through natural, conversational language and enables the robot to generalize beyond the explicit instructions given.\n\nLed by Andreea Bobu, assistant professor of aeronautics and astronautics at MIT, the team created FLEX to help robots reason and adapt in real-time based on a person's natural, conversational commands. This is particularly crucial for assistive and household robots that must carry out physical tasks and work in complex environments.\n\nThe FLEX platform works in a three-step process. In the \"Add\" step, a robot interprets a user's natural language command to add specific features to its list of objects it should avoid in the environment. For example, if a robot is moving a bowling ball and a person says \"Don't drop it on my mug,\" the robot adds the mug to its list of features to avoid.\n\nThe \"Generalize\" step, an innovative aspect of FLEX, allows the robot to use a large language model (LLM), such as GPT-4o, to identify patterns and connections among the specified features. By leveraging the vast amount of data the LLM has been trained on, it can make generalizations about avoiding objects based on their inherent properties, like fragility, even when the person has not explicitly stated it.\n\nFinally, in the \"Mask\" step, the robot's standard computer vision system identifies objects no longer present in the environment and removes them from the features list. This entire three-step process takes approximately five seconds and significantly enhances a robot's ability to interpret and respond to human commands in a more intuitive and flexible manner.",
  "summary": "The new approach, known as FLEX, could lead to more intuitive, interactive household and assistive robots.",
  "key_points": [
    "FLEX platform developed by MIT engineers",
    "Enables robots to understand human language beyond literal commands",
    "Three-step process: Add, Generalize, Mask"
  ],
  "editors_take": null,
  "illustration": null,
  "coverage": {
    "outlets": 1,
    "also_reported_by": []
  },
  "ai_generated": true,
  "disclaimer": "Summaries, key points and the editor’s take are written by software from other outlets’ reporting and may contain errors — always check the linked original."
}