PrimeSim: A Generalist World Model Simulator from Large-Scale Cross-Embodiment Demonstrations
PrimeSim: A Generalist World Model Simulator from Large-Scale Cross-Embodiment Demonstrations
This paper introduces PrimeSim, an action-conditioned, multiview world model simulator designed to faithfully follow commanded actions and maintain consistency across camera views.
Source: arXiv Robotics — research abstracts · Read original article ↗
Loading article text…
What the source reports
Publisher-reported claims, with original evidence. These results have not been independently verified by RoboSignal.
Reported numbers
training_hours
30,110 hours
View original evidence
raining. Starting from a pretrained video generator, we midtrain on 30,110 hours of robot demonstrations spanning 22 platforms. To improve action following, we introduce a unified cross-embodiment action representation and frame-aligned dual-path conditioning to strengthen action-frame correspondence. To maintain multiview consistency, we develop view-aware interleaved attention with explicit camera identities, enabl
Open source S3gradient_steps
100–200
View original evidence
ts coherent long-horizon simulation. PrimeSim runs at 10.1 FPS, remains stable over long-horizon rollouts on rigid and deformable manipulation tasks, and transfers to held-out platforms within 100 to 200 gradient steps where the pretrained backbone does not. Against five open-source action-conditioned world models it leads four of the five WorldSimProbe suites and posts the best overall EWMBench score, while fine-gra
Open source S5fps
10.1
View original evidence
ts coherent long-horizon simulation. PrimeSim runs at 10.1 FPS, remains stable over long-horizon rollouts on rigid and deformable manipulation tasks, and transfers to held-out platforms within 100 to 200 gradient steps where the pretrained backbone does not. Against five open-source action-conditioned world models it leads four of the five WorldSimProbe suites and posts the best overall EWMBench score, while fine-gra
Open source S5benchmarks
4
View original evidence
ts coherent long-horizon simulation. PrimeSim runs at 10.1 FPS, remains stable over long-horizon rollouts on rigid and deformable manipulation tasks, and transfers to held-out platforms within 100 to 200 gradient steps where the pretrained backbone does not. Against five open-source action-conditioned world models it leads four of the five WorldSimProbe suites and posts the best overall EWMBench score, while fine-gra
Open source S5
- paper: not_reported
- code: Reported pending Open source S6
- weights: Reported pending Open source S6
- dataset: not_reported
- code: not_reported
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
raining. Starting from a pretrained video generator, we midtrain on 30,110 hours of robot demonstrations spanning 22 platforms. To improve action following, we introduce a unified cross-embodiment action representation and frame-aligned dual-path conditioning to strengthen action-frame correspondence. To maintain multiview consistency, we develop view-aware interleaved attention with explicit camera identities, enabl
Open source S3
ts coherent long-horizon simulation. PrimeSim runs at 10.1 FPS, remains stable over long-horizon rollouts on rigid and deformable manipulation tasks, and transfers to held-out platforms within 100 to 200 gradient steps where the pretrained backbone does not. Against five open-source action-conditioned world models it leads four of the five WorldSimProbe suites and posts the best overall EWMBench score, while fine-gra
Open source S5
ined local calibration still trails the field; its rollouts are faithful enough to support GRPO-based policy optimization. We release the model weights, inference code, and training code.
Open source S6
Source：arXiv Robotics — research abstracts · arxiv.org