analiza aiqbench
This research introduces ActMem-VLA, a novel architecture for vision-language-action models that integrates explicit interaction history to mitigate perceptual aliasing in long-horizon robotic tasks. By employing a dual-expert design with a Mamba-based memory module and a lightweight steering expert, the method achieves significant performance gains while keeping the base policy frozen to minimize training costs. The approach effectively balances high-level task progression with fine-grained action refinement, offering a scalable solution for improving robot reliability in complex environments.
Remember What You Did: Action-History Memory with Dual-Expert Denoising for Long-Horizon Vision-Language-Action Policies
arXiv:2609.37307v1 Announce Type: new Abstract: Vision-language-action (VLA) models have driven rapid progress in robotic manipulation, demonstrating strong fine-grained control and promising performance on long-horizon tasks. However, many existing VLAs lack explicit access to interaction history, making them vulnerable to perceptual aliasing: similar current observations and robot states at different task stages may induce action ambiguity and lower success rate. Existing methods incorporate temporal or progress cues through feature conditioning, action-prior modification, or sampling guidance. However, methods that jointly fine-tune memory modules and the base VLA incur additional policy-training costs, motivating the separation of trainable history-conditioned steering from frozen base-policy refinement. We propose ActMem-VLA, a dual-expert handover architecture that augments a frozen, fine-tuned VLA with a memory plugin comprising a Mamba-based memory module and a lightweight PreAction Expert (PAE). Specifically, Mamba encodes executed-action history into memory that conditions PAE alongside current context. With these inputs, PAE steers task progression during early, high-noise denoising, then passes the partially denoised action to the frozen Action Expert (AE) to refine action details during the remaining low-noise steps. The fine-tuned base VLA remains frozen throughout training, while only the Mamba module and PAE are jointly optimized. On LIBERO-Mem, ActMem-VLA achieves 80.8\% average success across all ten tasks, compared with 65.2\% for $\pi_{0.5}$ and 49.5\% for MemoryVLA, while introducing only 3.45\% additional parameters. Across four real-world tasks, it improves the average success rate over $\pi_{0.5}$ by 28.8\%.
Abstract:Vision-language-action (VLA) models have driven rapid progress in robotic manipulation, demonstrating strong fine-grained control and promising performance on long-horizon tasks. However, many existing VLAs lack explicit access to interaction history, making them vulnerable to perceptual aliasing: similar current observations and robot states at different task stages may induce action ambiguity and lower success rate. Existing methods incorporate temporal or progress cues through feature conditioning, action-prior modification, or sampling guidance. However, methods that jointly fine-tune memory modules and the base VLA incur additional policy-training costs, motivating the separation of trainable history-conditioned steering from frozen base-policy refinement. We propose ActMem-VLA, a dual-expert handover architecture that augments a frozen, fine-tuned VLA with a memory plugin comprising a Mamba-based memory module and a lightweight PreAction Expert (PAE). Specifically, Mamba encodes executed-action history into memory that conditions PAE alongside current context. With these inputs, PAE steers task progression during early, high-noise denoising, then passes the partially denoised action to the frozen Action Expert (AE) to refine action details during the remaining low-noise steps. The fine-tuned base VLA remains frozen throughout training, while only the Mamba module and PAE are jointly optimized. On LIBERO-Mem, ActMem-VLA achieves 80.8\% average success across all ten tasks, compared with 65.2\% for $\pi_{0.5}$ and 49.5\% for MemoryVLA, while introducing only 3.45\% additional parameters. Across four real-world tasks, it improves the average success rate over $\pi_{0.5}$ by 28.8\%.
Robotics (cs.RO)
arXiv:2609.37307 [cs.RO]
(or arXiv:2609.37307v1 [cs.RO] for this version)
https://doi.org/10.48550/arXiv.2609.37307 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Text integral sub licența deschisă a publicației · 319 cuvinte
