LaST0.5: A Generalist World Action Model with Interaction-Centric Latent Reasoning
LaST0.5: A Generalist World Action Model with Interaction-Centric Latent Reasoning
This research presents LaST0.5, a novel World Action Model (WAM) designed to bridge the gap between global world modeling and precise robotic manipulation. The model introduces an interaction-centric latent space supervised by visual cues of robot-environment interactions, instantiated as a latent expert within a Mixture-of-Transformers architecture. By jointly predicting future representations and progress values, LaST0.5 adapts its world-to-action transfer using progress consistency to adjust action execution. Pretrained on over 14k hours of robotic data, the model achieves 99.5% success on the LIBERO benchmark and outperforms previous state-of-the-art methods on ten RLBench tasks, demonstrating robust long-horizon execution and dexterous manipulation in real-world experiments.
The paper introduces LaST0.5, a World Action Model that incorporates an interaction-centric latent reasoning module to better align world evolution with robotic manipulation tasks.
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
>14,000 hours
View original evidence
just action execution and trigger future resampling. Finally, we pretrain LaST0.5 on a large-scale robotic data corpus comprising over 14k hours of data, establishing interaction-centric physical reasoning as a scalable and effective paradigm. LaST0.5 achieves 99.5% on LIBERO, surpasses the previous state of the art on ten evaluated RLBench tasks, and demonstrates robust long-horizon execution and precise dexterous m
Open source S5tasks_evaluated
10
Reported trials
View original evidence
just action execution and trigger future resampling. Finally, we pretrain LaST0.5 on a large-scale robotic data corpus comprising over 14k hours of data, establishing interaction-centric physical reasoning as a scalable and effective paradigm. LaST0.5 achieves 99.5% on LIBERO, surpasses the previous state of the art on ten evaluated RLBench tasks, and demonstrates robust long-horizon execution and precise dexterous m
Open source S5
- paper: not_reported
- code: not_reported
- weights: not_reported
- dataset: not_reported
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
just action execution and trigger future resampling. Finally, we pretrain LaST0.5 on a large-scale robotic data corpus comprising over 14k hours of data, establishing interaction-centric physical reasoning as a scalable and effective paradigm. LaST0.5 achieves 99.5% on LIBERO, surpasses the previous state of the art on ten evaluated RLBench tasks, and demonstrates robust long-horizon execution and precise dexterous m
Open source S5
Source：arXiv Robotics — research abstracts · arxiv.org