Reproduce “HiSTEVE 1.0: Harnessing Human-Interface VLA Agents for Open-World Games” 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
Zecong Tang, Zian Guan, Yuncheng Guo, Zhanqiu Zhang, Yiwen Guo, Bo Dai
HiSTEVE organizes long-horizon Minecraft tasks by pairing a visual controller with a planner that selects feasible subtasks and verifies progress before continuing, recovering, or replanning. The controller uses first-person visual input and keyboard-and-mouse control. In a fixed-controller comparison, the full harness raises LongBench final-goal success from 10.7% to 25.0%, while atomic scores change little; recovery and open-ended tasks remain difficult.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.