Blender-generated datasets teach world models object permanence at 10k samples per task
_akhaliq · x · 2026-09-26
The paper "Training Object Permanence in World Models" (arXiv:2609.28654) argues object permanence — a foundation of human cognition — is missing from current video and world models. The authors build a data infrastructure of diverse object-permanence cognitive tasks, each with a Blender-based generator scalable to at least 10,000 diverse samples per task. They show this infrastructure effectively trains video models and world models that acquire object permanence, offering a systematic path toward human-like physical common sense in world models.
More from Research
- Cadence: An Architecture for a Continuous, Efficient Stream of Intelligence — rvp · 2026-09-26
- Dev trains a 50MB model in 15 min to replace Gemini Flash at 0.06s latency — newz2000 · 2026-09-26
- WetLabs Benchmark tests robots on 9 wet-lab tasks: can machines accelerate science? — ericjang11 · 2026-09-26
- Medmarks v1.0 lands NeurIPS track: medical LLM benchmark now covers 30 suites, 61 models — iScienceLuvr · 2026-09-26
- AI-driven lab finds iridium-free palladium catalyst InMnPdOx stable for 1,000 hours — CatAstro_Piyush · 2026-09-26
- Agate-001-preview: 260M-param open-weights image model nears SD 1.5 with thinker-renderer architecture — incorporo · 2026-09-26