{"version":1,"type":"story","url":"https://digestai.news/story/researchers-introduce-mirror-a-selfsupervised-videotoaction-model","json":"https://digestai.news/story/researchers-introduce-mirror-a-selfsupervised-videotoaction-model.json","markdown":"https://digestai.news/story/researchers-introduce-mirror-a-selfsupervised-videotoaction-model.md","slug":"researchers-introduce-mirror-a-selfsupervised-videotoaction-model","headline":"Researchers introduce Mirror, a self‑supervised video‑to‑action model","summary":"The paper presents Mirror, a self‑supervised approach that learns to map video plans to actions without any pre‑collected action data or hand‑crafted rewards. Mirror iteratively tries to follow a visual plan, records the resulting video‑action pairs, and uses them to improve its own predictions, gradually closing the gap to the visual planner.\n\nExperiments span Atari games, OGBench benchmarks, DeepMind Control suites, and the Minecraft BASALT challenge. Mirror outperforms leading learning‑from‑observation methods when paired with learned video plans and reaches near‑expert performance with oracle plans. In Minecraft survival, it is the first method to acquire an iron pickaxe solely through self‑improvement, and in the offline setting it sets a new state‑of‑the‑art result for diamond acquisition using expert labels only.\n\nThe results suggest that advances in video generative models can be leveraged to build increasingly capable embodied agents without relying on large expert action datasets or manually designed reward functions.","keyPoints":["Mirror learns video‑to‑action mappings through self‑supervised interaction, needing no prior action data","Across Atari, OGBench, DeepMind Control and Minecraft, Mirror outperforms leading observation‑learning methods","In Minecraft survival, Mirror self‑improves to acquire an iron pickaxe without rewards or expert labels"],"whyItMatters":"Shows that video generative models can power autonomous agents without costly data collection, lowering barriers for scalable embodied AI across games and simulated environments.","category":{"slug":"agents","name":"Agents & Tools","url":"https://digestai.news/category/agents"},"entities":{"companies":[],"models":["Mirror"],"people":[]},"firstPublishedAt":"2026-10-20T04:00:00Z","updatedAt":"2026-10-20T04:00:00Z","sourceCount":1,"hasPrimarySource":true,"sources":[{"outlet":"arXiv cs.AI","title":"Self-Improving Video-to-Action Models","url":"https://arxiv.org/abs/2610.24177","publishedAt":"2026-10-20T04:00:00Z","type":"primary","primary":true,"lead":true}],"sourceNotes":null,"discussions":[],"thread":null,"cite":{"text":"Digest AI, \"Researchers introduce Mirror, a self‑supervised video‑to‑action model\", 20 October 2026, https://digestai.news/story/researchers-introduce-mirror-a-selfsupervised-videotoaction-model","publisher":"Digest AI","title":"Researchers introduce Mirror, a self‑supervised video‑to‑action model","datePublished":"2026-10-20T04:00:00Z","url":"https://digestai.news/story/researchers-introduce-mirror-a-selfsupervised-videotoaction-model"},"generatedBy":"Written by Digest AI's editorial model from the linked sources; the sources are the record.","license":"Headlines, digests and key points are written by Digest AI and may be quoted with a link to the story page. Linked articles belong to their publishers. Terms: https://digestai.news/terms#reuse"}