Reproduce “Decision Transformer with Task-Aware Representations for Offline Meta-Reinforcement Learning” 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
Xinbo Zhao, Yuhang Liu, Xin Zhang, Lei Le, Zhaohan Xi, Yingxue Zhang
CoTA-DT combines trajectory, text, and predicted reward and dynamics signals to form a task representation that conditions a Decision Transformer. For zero-shot use, it initializes the latent from a task description and can update it using rollout history. On five held-out Meta-World tasks with Mixed offline data, it reports return 1292.5 ± 28.2 versus 1103.5 ± 36.3 for T2DA-T, within this benchmark setting.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.