SMAX

该数据集名为SMAX，是流行的星Craft多智能体挑战的简化版，它经过了向量化处理，旨在实现GPU加速，并提供一个更加灵活的多智能体强化学习环境。SMAX支持自我对弈、元学习以及在多智能体强化学习中的其他未来应用，与传统实现方式相比，它能显著提高训练速度。该数据集还能支持多个团队的训练，单个训练运行中最多可达512个团队。其任务是进行多智能体强化学习。This dataset, named SMAX, is a simplified variant of the well-known StarCraft Multi-Agent Challenge. It has been vectorized to support GPU acceleration and offers a more flexible multi-agent reinforcement learning environment. SMAX enables self-play, meta-learning, and other potential future applications in multi-agent reinforcement learning, and drastically boosts training speed relative to traditional implementations. Additionally, it supports training across multiple teams, with a maximum of 512 teams permitted in a single training run. This dataset is tailored for multi-agent reinforcement learning tasks.