MAgnet

该数据集包含了一系列具有非单调收益的合作机器人任务挑战，包括提升、三倍提升、追捕和老虎等任务，这些任务旨在评估多代理强化学习方法。在这些任务中，要求代理采取最优的合作行动，而对次优行动则会施加惩罚，这影响了学习动态。数据集规模多样，涵盖了不同复杂度和代理需求的多任务。总体而言，这是一个在非单调环境中进行合作多代理强化学习的任务。

This dataset contains a series of cooperative robotics task challenges with non-monotonic rewards, including lifting, triple-lifting, pursuit, and the classic tiger problem. These tasks are designed to evaluate multi-agent reinforcement learning methods. In these tasks, agents are required to take optimal cooperative actions, while suboptimal actions are penalized, which affects the learning dynamics. The dataset has diverse scales, covering multiple tasks with varying complexities and agent requirements. Overall, this dataset serves as a benchmark for cooperative multi-agent reinforcement learning research in non-monotonic environments.