Faster-WAM: Efficient Inference-Time Future Conditioning for Robust World Action Models
Quick Answer
Faster-WAM introduces an efficient future-conditioning framework for World Action Models, enhancing robustness in robot manipulation.
Quick Take
It achieves a 49.14% to 73.57% success rate improvement on the LIBERO-Plus benchmark while running 2.21x faster than Joint-WAM, demonstrating superior performance and efficiency.
Key Points
- Faster-WAM preserves future representations while avoiding costly video-action interactions.
- Introduces SparseMoT for selective video-action interaction at fewer network stages.
- Interval KV-Fusion aggregates multi-depth future representations without increasing attention complexity.
- Achieves state-of-the-art performance on LIBERO and RoboTwin 2.0 benchmarks.
- Demonstrates strong robustness in real-world manipulation tasks.
DeepSignal Analysis
What happened
Faster-WAM is a new framework designed to enhance the efficiency of World Action Models (WAMs) in robot manipulation. It maintains future representations during inference while improving performance and speed, achieving a success rate increase on the LIBERO-Plus benchmark from 49.14% to 73.57% and running 2.21 times faster than Joint-WAM.
Key evidence
- Faster-WAM improves the success rate on the LIBERO-Plus benchmark from 49.14% to 73.57%, indicating a significant enhancement in performance.
- The framework operates 2.21 times faster than Joint-WAM, suggesting improved computational efficiency.
- Faster-WAM achieves state-of-the-art performance on LIBERO and RoboTwin 2.0, demonstrating its robustness in real-world manipulation tasks.
Why it matters
The development of Faster-WAM addresses a critical challenge in robot manipulation by balancing the need for future-aware representations with computational efficiency. This is particularly relevant as robots increasingly operate in dynamic environments where adaptability is crucial. The improvements in success rates and speed could lead to more effective and reliable robotic systems in practical applications.
Paper Resources
📖 Reader Mode
~2 min readAbstract:World Action Models (WAMs) improve robot manipulation by learning how the environment evolves beyond the current observation. However, existing approaches face a fundamental dilemma: Joint-WAMs preserve future-aware representations during inference but incur prohibitive computation costs, while efficient alternatives remove future modeling at inference time and may lose the robustness benefits of temporal reasoning. In this work, we revisit the role of future representations in WAMs and show that inference-time future conditioning is critical for generalization under distribution shifts. This observation motivates Faster-WAM, an efficient future-conditioning WAM that preserves future representations while avoiding expensive video-action interaction. Faster-WAM introduces a sparse future-conditioning framework that computes future representations once and selectively reuses them throughout action denoising. Specifically, we propose SparseMoT to replace ubiquitous layer-wise fusion with selective video-action interaction at a compact subset of network stages, and Interval KV-Fusion to aggregate multi-depth future representations without increasing attention complexity. Experiments demonstrate that Faster-WAM achieves a substantially better performance-efficiency trade-off than existing WAMs. On the out-of-distribution LIBERO-Plus benchmark, Faster-WAM improves success rate from 49.14% to 73.57% compared with Fast-WAM, while running 2.21$\times$ faster than Joint-WAM. It further achieves state-of-the-art performance on LIBERO and RoboTwin 2.0, while demonstrating strong robustness in real-world manipulation.
| Subjects: | Computer Vision and Pattern Recognition (cs.CV) |
| Cite as: | arXiv:2608.04404 [cs.CV] |
| (or arXiv:2608.04404v1 [cs.CV] for this version) | |
| https://doi.org/10.48550/arXiv.2608.04404 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Weiheng Zhao [view email]
[v1]
Wed, 5 Aug 2026 03:18:58 UTC (2,910 KB)
— Originally published at arxiv.org
Want this in your inbox every morning?
Daily brief at your local 8am — bilingual EN/中文, free.
More from arXiv cs.CV
See more →ProMoE-FL: Prototype-conditioned Mixture of Experts for Multimodal Federated Learning with Missing Modalities
ProMoE-FL introduces a Prototype-conditioned Mixture-of-Experts framework for multimodal federated learning, effectively addressing missing modalities. It outperforms existing methods on four chest X-ray datasets, demonstrating superior feature synthesis capabilities in both homogeneous and heterogeneous settings.