Reproduce “MemBodied: Recurrent Associative Memory for Vision-Language-Action Models” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Tej Deep Pala, Navonil Majumder, Bryce Goh, Raphael Yee, Jianfei Yang, Liming Chen, Soujanya Poria
MemBodied gives a robot fixed associative memory and an initial-scene anchor instead of appending past observations to each policy input. It updates layer-wise matrices with actions and observed consequences. Across five RMBench tasks, it achieved 50.0% mean success versus 6.4% for stateless π₀, or 7.81× that baseline, supporting memory on tested history-dependent tasks, not universal reliable success.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.