Reproduce “Value-Decomposition Multi-Agent Actor-Critics” 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
Jianyu Su, Stephen Adams, Peter A. Beling
VDAC trains decentralized policies for StarCraft II micromanagement. Agents estimate local state values, combined monotonically by a centralized critic using global state information; execution uses local observation histories. On SMAC’s 3s5z scenario, VDAC-mix’s median test win rate at 2 million training steps exceeded that of QMIX trained under A2C by 71%; this finding is specific to that map and training setup.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.