Video planners beat world models in Minecraft: 2x on iron pickaxe, 8x on diamonds
vincesitzmann · x · 2026-10-21
Vincent Sitzmann and Hyunwoo Ryu argue video planning is the better path over action-conditioned world models: in offline Minecraft, their video planner + IDM significantly outperforms Dreamer 4, the SOTA world-model agent — roughly 2x success on iron pickaxes and 8x on diamonds. The core argument: data is the deciding factor — there's abundant video on the internet, so training a video planner (their Vera / mirror models) is the easier path to generality, even when expert demonstrations include action labels, since action labels are scarce. Any policy that can't scale with diverse data at scale is hopeless.
More from Embodied
- 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
- Dream-1 controlled 31 unseen robot types across 53 countries within 24 hours — gharik · 2026-10-21