analiza aiqbench
HiMem-WAM addresses a critical gap in World Action Models by integrating hierarchical latent actions with boundary-triggered memory updates, enabling robust long-horizon robotic manipulation without relying on test-time video generation. The approach demonstrates that structured temporal abstraction significantly improves generalization under deployment perturbations while effectively handling memory-dependent tasks. This work advances embodied intelligence by providing a scalable framework for causal inference in complex physical environments.
HiMem-WAM: Hierarchical Memory-Gated World Action Models for Robotic Manipulation
arXiv:2606.10363v1 Announce Type: new Abstract: World Action Models (WAMs) have emerged as a new powerful paradigm for embodied intelligence, learning action-relevant visual dynamics that significantly enhance generalization and robustness. However, existing WAMs still struggle with task-relevant memory in long-horizon robotic manipulation. To address this, we present HiMem-WAM, a Hierarchical Memory-Gated WAM that integrates motion-centric latent actions, high-level skill latents, and boundary-triggered memory updates. Specifically, we develop a hierarchical latent action framework that jointly learns low-level motion and high-level skill latents, providing structured temporal abstraction. Meanwhile, a boundary-aware memory gate writes compact task states at predicted skill transitions, enabling causal inference without test-time generation of future video or optical flow estimation. Evaluated on LIBERO, LIBERO-PLUS, RMBench and real-world tasks, HiMem-WAM shows that hierarchical latents improve robustness under deployment perturbations, and the memory module substantially benefits memory-dependent long-horizon manipulation.
Abstract:World Action Models (WAMs) have emerged as a new powerful paradigm for embodied intelligence, learning action-relevant visual dynamics that significantly enhance generalization and robustness. However, existing WAMs still struggle with task-relevant memory in long-horizon robotic manipulation. To address this, we present HiMem-WAM, a Hierarchical Memory-Gated WAM that integrates motion-centric latent actions, high-level skill latents, and boundary-triggered memory updates. Specifically, we develop a hierarchical latent action framework that jointly learns low-level motion and high-level skill latents, providing structured temporal abstraction. Meanwhile, a boundary-aware memory gate writes compact task states at predicted skill transitions, enabling causal inference without test-time generation of future video or optical flow estimation. Evaluated on LIBERO, LIBERO-PLUS, RMBench and real-world tasks, HiMem-WAM shows that hierarchical latents improve robustness under deployment perturbations, and the memory module substantially benefits memory-dependent long-horizon manipulation.
Robotics (cs.RO)
arXiv:2606.10363 [cs.RO]
(or arXiv:2606.10363v1 [cs.RO] for this version)
https://doi.org/10.48550/arXiv.2606.10363 arXiv-issued DOI via DataCite (pending registration)
Text integral sub licența deschisă a publicației · 149 cuvinte
Subiecte conexe
Build a Physical AI model factory with NVIDIA Cosmos 3 on SageMaker HyperPod
AWS · 4 sept. 2026
RM-WAM: Reliable Memory-Augmented World Action Models for Long-Horizon Robot Manipulation
arXiv · 14 oct. 2026
World Models and Yann LeCun’s Bet Against Large Language Models
Quantum Zeitgeist · 17 iul. 2026
PileBelief: Persistent Physical State for Interaction-Driven World Modeling
arXiv · 22 sept. 2026