Reproduce “MemRole: From Memories to Decisions for Long-Horizon Reasoning in Role-Playing Agents” 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
Mengting Li, Xilong Cheng, Zhihao Xiao, Zhenhua Xu, Jingyuan Zhou, Linfeng Li, Long Zhang, Borui Cai and 1 more
MemRole trains role-playing agents to turn past experiences into decisions for the present scene. Its three-stage process interprets the interaction, activates and weighs relevant memories, then plans a response; supervised traces and critique-guided self-distillation train this process. On MemRole Test, MemRole-32B scores 64.25 versus 49.05 for its Qwen3-32B backbone, but does not outperform every frontier model.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.