Google Research Football (Gfootball)

该数据集是一个双玩家足球（足球）环境，每个智能体控制一个由11名球员组成的球队。状态信息包括球员和球的各种位置数据，这为复杂的强化学习场景提供了可能。此外，该数据集包含了用于对抗性训练的各种配置，包括黑盒和白盒设置，并采用近端策略优化（Proximal Policy Optimization）进行训练。观测数据的规模为72 x 96 x 16像素。任务旨在训练强化学习智能体，使其能够相互对抗，发展出对抗性策略。

This dataset is a two-player soccer (football) environment where each agent controls a team of 11 players. The state information encompasses various positional data of players and the ball, enabling the construction of complex reinforcement learning scenarios. Additionally, this dataset contains various configurations for adversarial training, including black-box and white-box settings, and is trained using Proximal Policy Optimization (PPO). The observational data has a size of 72 × 96 × 16 pixels. The task aims to train reinforcement learning agents to compete against each other and develop adversarial strategies.