UniWALM: Unified World Action Language Model for Embodied Robot Control
UniWALM: Unified World Action Language Model for Embodied Robot Control
UniWALM is a unified world-action-language model that learns video, action, and language in a shared representation, improving manipulation and memory-dependent behavior. It unifies all modalities in a single transformer backbone, with diffusion heads that co-denoise synchronously. The model achieves 34.5% performance on memory-dependent tasks without pretrained language models and improves success rates in real-world tasks from 11.1% to 61.1% by generating textual memory.
UniWALM introduces a unified world-action-language model that integrates video, action, and language into a shared representation, enabling memory-dependent behavior and improved manipulation tasks. The model closes a language-as-memory loop without external planners, and demonstrates significant performance gains in real-world tasks compared to existing methods.
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.
- paper: not_reported
- code: not_reported
- weights: not_reported
- dataset: not_reported
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
s over same-backbone baselines (e.g., FastWAM) on compositional pick-and-place (MolmoSpaces) and across the memory-dependent subset of RMBench, where it reaches 34.5% without any pretrained language or reasoning model and exceeds Mem-0's keyframe-equipped setup (28.5%) using only a single anchor frame. On a real-world Find-the-Item task, generating textual memory lifts success from 11.1% to 61.1%, while state-of-the-
Open source S5
art VLAs stay at or below chance (<=33.3%).
Open source S6
Source：arXiv Robotics — research abstracts · arxiv.org