Bedrock-RL：用于训练和评估 VLM 智能体的确定性 Minecraft 框架
原标题：Bedrock-RL: a deterministic Minecraft framework for training and benchmarking VLM agents
AI 摘要
Hugging Face 博客发布了 bedrock-rl，一个用于训练和评估视觉语言智能体（VLM agents）的确定性 Minecraft 框架。它结合了 Netherite（Minecraft 模拟的确定性 C/CUDA 重实现）和 verl（分布式强化学习库），支持多种训练方法（如 GRPO、SFT 等），并采用可替换组件的设计，通过 YAML 配置定义任务和奖励。示例显示 Qwen3-VL 2B 在 GRPO 训练后，pass@3 成功率从 10.8% 提升至 22.2%。该框架旨在提供可复现、可扩展的基准测试环境。
正文节选
Minecraft is close to an ideal testbed for embodied vision-language agents: open-ended, spatial, and full of long-horizon tasks with programmatically verifiable outcomes. Historically, however, it has been a challenging environment to optimize RL in. The Java client is slow, non-deterministic, and hostile to parallel rollouts, so most prior work settled for behavior cloning on human video. bedrock-rl fixes the plumbing. It couples Netherite, a deterministic C/CUDA reimplementation of the Minecra