{"rewrite":{"id":"r_5b9c0e10694372e6b26fd884","clusterId":"c_06cf9247e54aa909edcc30a8","slug":"nvidia-develops-enpire-framework-to-let-ai-agents-improve-robot-skills-autonomously","model":"deepseek-v4-flash","headline":"Nvidia Develops ENPIRE Framework to Let AI Agents Improve Robot Skills Autonomously","summary":"Nvidia announced ENPIRE, a harness framework developed with Carnegie Mellon University and UC Berkeley that lets AI agents like Claude Code and Codex autonomously improve real-world robot task execution. In tests, it achieved a 99% success rate on tasks like inserting a pin and cutting a cable tie by running a continuous improvement loop.","whyItMatters":"ENPIRE addresses a key bottleneck in robotics-the need for human supervision and algorithm design-by enabling AI agents to iteratively refine robot behavior in the physical world without direct human intervention.","webCardHtml":"\u003cp\u003eNvidia has unveiled ENPIRE, a harness framework that bridges AI agents and real-world robots. Developed with Carnegie Mellon University and UC Berkeley, ENPIRE lets agents such as Claude Code, Codex, and Kimi Code autonomously develop, test, and improve robot algorithms. In experiments, the framework boosted task success rates to 99% for actions like inserting a pin, setting a GPU on a board, and cutting a cable tie. The system runs a loop: execute a task, plan improvements based on results, update the algorithm, and apply it to the robot. Nvidia notes that running more robots simultaneously speeds up improvement, though it also increases GPU waste and token consumption.\u003c/p\u003e","blueskyPost":"ENPIRE's 99% success rate on insertion and cable-cutting tasks suggests the loop, not the agent, is the breakthrough. Claude Code and Codex are interchangeable in this harness.","twitterPost":"ENPIRE's 99% success rate on insertion and cable-cutting tasks suggests the improvement loop, not the agent, is the breakthrough.","threadsPost":"ENPIRE's 99% success rate on tasks like pin insertion and cable cutting suggests the continuous improvement loop, not the specific AI agent, is the real advance. Claude Code and Codex are interchangeable in this harness.","newsletterBlurb":"Nvidia announced ENPIRE, a harness framework that lets AI agents like Claude Code autonomously improve real-world robot performance. Developed with Carnegie Mellon and UC Berkeley, it achieved 99% success on tasks such as inserting a pin and cutting a cable tie. The system runs a continuous improvement loop, but Nvidia notes it can waste GPU resources and increase token costs.","attributionJson":"[{\"source\":\"GIGAZINE\",\"url\":\"https://gigazine.net/news/20260619-nvidia-enpire-agentic-robot/\",\"title\":\"NVIDIA develops 'ENPIRE,' a mechanism to autonomously improve robots with AI agents like Claude Code\"}]","lintFlagsJson":null,"lintHits":0,"costUsd":0,"inputTokens":3892,"outputTokens":555,"status":"published","repairAttempts":0,"nextRepairAt":null,"factsAttemptedAt":1781841930,"createdAt":"2026-06-19T04:00:21.000Z","publishedAt":"2026-06-19T04:03:16.000Z","updatedAt":"2026-06-19T04:03:16.000Z"},"cluster":{"id":"c_06cf9247e54aa909edcc30a8","canonicalTitle":"Claude CodeなどのAIエージェントでロボットを自律的に改善する仕組み「ENPIRE」がNVIDIAによって開発される","representativeArticleId":"a_5c970f2bb697eb22315d138d","sourceCount":1,"writtenSourceCount":1,"writeAttempts":0,"isSolo":true,"entitiesJson":"{\"anime_titles\":[],\"manga_titles\":[],\"work_titles\":[\"ENPIRE\"],\"studios\":[\"NVIDIA\"],\"people\":[],\"type\":\"announcement\",\"domain\":\"other\",\"is_roundup\":false}","contentType":"news","status":"published","firstSeenAt":"2026-06-19T03:24:00.000Z","lastSeenAt":"2026-06-19T03:24:00.000Z","updatedAt":"2026-06-19T04:03:18.000Z"},"attribution":[{"source":"GIGAZINE","url":"https://gigazine.net/news/20260619-nvidia-enpire-agentic-robot/","title":"Claude CodeなどのAIエージェントでロボットを自律的に改善する仕組み「ENPIRE」がNVIDIAによって開発される"}],"entities":{"anime_titles":[],"manga_titles":[],"work_titles":["ENPIRE"],"studios":["NVIDIA"],"people":[],"type":"announcement","domain":"other","is_roundup":false},"keyFacts":["Nvidia announced ENPIRE, a framework developed with Carnegie Mellon University and UC Berkeley that lets AI agents like Claude Code and Codex autonomously improve real-world robot task execution.","In experiments, ENPIRE achieved a 99% success rate on tasks including inserting a pin, setting a GPU on a board, and cutting a cable tie.","ENPIRE runs a continuous improvement loop: execute a task, plan improvements based on results, update the algorithm, and apply it to the robot.","Nvidia said running more robots simultaneously speeds up improvement but increases GPU waste and token consumption."]}