PrimeSim: world model trained on 30k hours of robot demos runs near real-time on one A100
jiqizhixin · x · 2026-11-15
PrimeBot Research introduces PrimeSim, an action-conditioned world model (arXiv preprint) that can improve robot policies later evaluated in physics simulators.
- Trained from a video generator on 30,110 hours of demos across 22 robot platforms, using a shared action representation with frame-aligned conditioning and view-aware attention for multi-camera input.
- Post-training adds causal generation, four-step sampling, training on its own rollouts, and hierarchical memory; on a single A100-80GB the reported setup generates 10.1 frames/sec across three 288×384 views (config-specific throughput).
- GRPO training inside PrimeSim reportedly raises the OpenVLA-OFT policy's LIBERO success rate.
More from Embodied
- RaiSim vs Newton benchmark: CPU robot sim beats GPU by up to 95x at practical batch sizes — ChongZzZhang · 2026-11-15
- Researcher teaches a Unitree G1 humanoid to ride a mechanical bull after CoRL — RemiCadene · 2026-11-15
- Viral claim: Claude runs a robot arm on a real factory task with zero training — but the model name doesn't exist — IgorCarron · 2026-11-15
- NEO-VLA ditches visual encoders for robot actions, hitting 90.9% on LIBERO frozen — 机器之心 · 2026-11-15
- Zipline's Keller Cliffton on 3M drone deliveries and what breaks at 10x scale — a16z Podcast · 2026-11-15
- AI Tamagotchi: CMU student's E Ink gadgets turn objects into needy pets — Wired AI · 2026-11-15