ABot-M0.5: Unified Mobility-and-Manipulation World Action Model

2026-07-01

AI Summary

Existing embodied learning approaches struggle with fine-grained contact dynamics, action conflicts, and long-horizon error accumulation in mobile manipulation tasks due to the absence of explicit world modeling. This work proposes a unified world-action model that introduces latent intermediate actions to bridge visual states and control signals, and employs a dual-level Mixture-of-Transformers architecture to decouple navigation from manipulation. Through a tripartite alignment strategy—temporal granularity alignment, action space disentanglement, and training-inference consistency—combined with a dreamer-forced training scheme, the method significantly enhances the robustness and accuracy of long-horizon reasoning. It achieves state-of-the-art performance on challenging benchmarks involving complex locomotion and dexterous manipulation, substantially improving both task success rates and control precision.

Abstract

Mobile manipulation is a key capability for general-purpose robots, yet remains challenging for current embodied learning methods. VLA policies are typically reactive and lack explicit world modeling, while existing World Action Models (WAMs) are still poorly aligned with the structure of mobile manipulation: they operate on coarse video chunks, model entangled navigation-manipulation actions, and train inverse dynamics under supervision that does not match autoregressive inference. As a result, they often miss fine-grained contact dynamics, suffer from action-distribution conflicts, and accumulate errors over long-horizon rollouts. We propose ABot-M0.5, a new WAM built on the insight that mobile manipulation requires alignment at three levels: temporal granularity, action space, and train-test consistency. To align temporal granularity, we introduce intermediate latent actions that capture local visual state transitions and serve as an bridging action space between video latents and embodiment-specific controls. To align action space, we design a dual-level Mixture-of-Transformers architecture that disentangles both modality representations and heterogeneous action subspaces such as base movement and arm manipulation. To align inference conditions, we propose the dream-forcing training strategy that progressively trains inverse dynamics on model-predicted videos, improving train-test alignment and robustness during autoregressive prediction. Experiments on challenging mobile and fine-grained manipulation benchmarks demonstrate that ABot-M0.5 achieves state-of-the-art performance in both long-horizon task success and finegrained control accuracy. These results highlight the critical importance of granularity-aligned, action-disentangled, and inference-consistent world-action modeling.

Problem

• mobile manipulation
• World Action Models
• action disentanglement
• temporal granularity
• train-test consistency

Innovation

• World Action Model
• mobile manipulation
• latent action
• Mixture-of-Transformers
• dream-forcing