Long-WAM: Scaling the Context of World-Action Models

Wei Huang, Bohan Zhang, Chenzhi Liu, Isabella Liu, Shuai Yang, Weian Mao, Luozhou Wang, Yicheng Xiao, Weifeng Lin, Qixin Hu, Bryan Chu, Sifei Liu, Linxi Fan, Xiaojuan Qi, Song Han, Yukang Chen

arXiv: 2610.10528Published: October 7, 202688 upvotes

View on arXiv Read PDF

AI summary

TL;DR

Long-WAM is a model-system framework that scales the context of causal world-action models for real-time robot control. It achieves state-of-the-art results on several benchmarks, including RoboCasa GR-1, LIBERO-Long, and DOMINO. Long-WAM enables dynamic and long-horizon manipulation with high success rates.

Why It Matters

Long-WAM's ability to process longer histories and adapt to real-time control constraints has significant implications for robotics and autonomous systems. It enables more accurate and efficient control, leading to improved performance in complex tasks. This has potential applications in areas like manufacturing, healthcare, and transportation.

Key Findings

Long-WAM outperforms existing methods on several benchmarks
Pretraining with autoregressive models improves performance
Long-WAM enables real-time deployment on various hardware platforms

Topics

autoregressive

robotics

realtime

control

multimodal

Thank you to arXiv for use of its open access interoperability. arxiv.org. aibrief.ai is not affiliated with or endorsed by arXiv.