Mirror hits Minecraft milestones via pure self-play, 9x DreamerV3's iron pickaxe rate without rewards
vincesitzmann · x · 2026-10-22
Mirror achieves Minecraft survival milestones using only self-play — no hand-crafted rewards or expert actions — reaching 9x DreamerV3's iron pickaxe success rate (5x under DreamerV3's own eval protocol). Unlike DreamerV3, which needed simplified mining actions because random exploration rarely mines a block, Mirror's agent persistently works on the mining milestone until it succeeds. With an oracle video planner it even finds diamonds, unmatched by online baselines at matched interaction time.
More from Embodied
- Unitree's New H2-X Humanoid Pictured Next to G1, Height Gap Revealed — micoolcho · 2026-10-22
- Explainer: How 3D Gaussian Splatting Gives Robots Spatial Memory — TheTuringPost · 2026-10-22
- Walden Robotics CEO Russ Tedrake on how general-purpose robots could reshape manufacturing — adnothing · 2026-10-22
- 3D-printed print mechanism snaps together with hooks and an alignment pin — _Stocko_ · 2026-10-22
- Open textbook explains when controller stability is topologically impossible — burkov · 2026-10-22
- Neuralink user controls robot vacuum with thoughts in a first-ever home demo — Scobleizer · 2026-10-22