UniWALM: Unified World Action Language Model for Embodied Robot Control
arXiv cs.AI · 21 Oct 2026, 03:59 AEST
UniWALM is a unified world-action-language model designed for embodied robot control, integrating video, action, and language into a single representation. It improves manipulation and memory-dependent behavior without the need for separate planning stages.
Key facts
- UniWALM unifies all three modalities inside a single mixture-of-transformers backbone whose per-modality diffusion heads co-denoise synchronously.
- The generated language is re-consumed on the next step, closing a language-as-memory loop without an external planner or memory bank.
- UniWALM consistently improves over same-backbone baselines (e.g., FastWAM) on compositional pick-and-place (MolmoSpaces) and across the memory-dependent subset of RMBench.
- On a real-world Find-the-Item task, generating textual memory lifts success from 11.1% to 61.1%, while state-of-the-art VLAs stay at or below chance (<=33.3%).
Read aloud
UniWALM is a unified world-action-language model designed for embodied robot control, integrating video, action, and language into a single representation. It improves manipulation and memory-dependent behavior without the need for separate planning stages. UniWALM unifies all three modalities inside a single mixture-of-transformers backbone whose per-modality diffusion heads co-denoise synchronously. The generated language is re-consumed on the next step, closing a language-as-memory loop without an external planner or memory bank. UniWALM consistently improves over same-backbone baselines e.g., FastWAM on compositional pick-and-place MolmoSpaces and across the memory-dependent subset of RMBench. On a real-world Find-the-Item task, generating textual memory lifts success from eleven point one percent to sixty one point one percent, while state-of-the-art VLAs stay at or below chance <=thirty three point three percent.
Sources in this cluster
- UniWALM: Unified World Action Language Model for Embodied Robot Control
arXiv cs.AI · 21 Oct 2026, 03:59 AEST