Reproduce “Test-Time Training as Residual Memory for Robot Policies” 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
Haoxuan Wang, Gengyu Zhang, Ramana Rao Kompella, Gaowen Liu, Yan Yan
TTT-RM combines a bounded memory bank of selected events with online updates to fast model weights. A history decoder estimates which past visual features current observations and retrieved records cannot reconstruct, and trains the weights to preserve that residual information. On real-world stowing, it reports 77.0% mean task progress versus 50.5% for the strongest baseline across 20 trials, not full-task success.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.