Mirror lets agents follow video-model plans, crafting an iron pickaxe in Minecraft
vincesitzmann · x · 2026-10-20
Mirror is a new framework where an agent learns to execute task plans imagined by video models, simply by practicing. As video models improve, agents improve with them — no extra action data needed. In Minecraft, Mirror crafted an iron pickaxe without rewards or action labels. The team argues that if video models scale like LLMs, they'll soon generate perfect video plans for any embodied task.
More from Embodied
- Amazon FAR's FIBER uses action-aligned latent planning to let Unitree G1 open doors and carry boxes — ChongZzZhang · 2026-10-21
- Sentdex trains robots to go from lying down to a half kneel, eyeing a full stand — Sentdex · 2026-10-21
- NVIDIA open-sources MotionBricks: 350K motion skills at 15,000 FPS, 2ms latency — aftahi_ai · 2026-10-21
- NVIDIA Cosmos3-H-Dreams: Generative Simulation Brings Clinical Surgery to Life — Hugging Face Blog · 2026-10-21
- FLoTS-SLAM uses optical-flow residuals to hit 0.73 DM-IoU in dynamic monocular SLAM — zhenjun_zhao · 2026-10-21
- Plato ring launches at $199 to turn near-silent speech into AI commands — TinfoilTricorn · 2026-10-21