BRAX

该数据集包含了源自原始Mujoco环境的连续控制运动任务，涵盖了蚂蚁、Walker2d、半猎豹和人形机器人等场景。这些环境的观测空间大小分别为87、17、18和227，而动作空间大小分别为8、6、6和17。此外，这些环境还配备了用于评估智能体性能的封装器。该数据集以高吞吐量、大规模并行化的环境为特色，其任务是在最大化前进进度和机器人稳定性的同时，最小化能量消耗。

This dataset contains continuous control locomotion tasks sourced from the original Mujoco environments, covering scenarios such as Ant, Walker2d, HalfCheetah, and Humanoid. The observation space sizes of these environments are 87, 17, 18, and 227 respectively, while their action space sizes are 8, 6, 6, and 17 respectively. Additionally, these environments are equipped with wrappers for evaluating agent performance. This dataset features high-throughput, large-scale parallelized environments, where the task is to minimize energy consumption while maximizing forward progress and robot stability.