SG6 benchmark

该数据集采用了SG6基准来评估在安全性至关重要的任务中，通过约束马尔可夫决策过程建模的强化学习算法的性能。该数据集包含64x64像素的RGB图像，并使用平均未折扣的回合回报和成本回报等指标来衡量性能。该数据集在多个环境中进行了规模扩展，包括PointGoal1、PointGoal2、CarGoal1、PointButton1、PointPush2和DoggoGoal1等环境。其任务是进行安全关键任务中的策略优化。

This dataset employs the SG6 benchmark to evaluate the performance of reinforcement learning algorithms modeled via constrained Markov decision processes in safety-critical tasks. It includes RGB images with a resolution of 64x64 pixels, and uses metrics such as average undiscounted episode return and cost return to assess performance. This dataset has been scaled across multiple environments, including PointGoal1, PointGoal2, CarGoal1, PointButton1, PointPush2, and DoggoGoal1. The core task of this dataset is policy optimization for safety-critical tasks.