JaxGCRL Benchmark

该数据集是一个为自我监督目标条件强化学习（GCRL）而设计的高性能代码库和基准测试，它使得智能体能够在多种环境中快速进行训练。该基准测试允许评估对比强化学习中的关键设计选择，并促进对各种能量函数和损失目标的有效测试。该数据集的规模包括1024个并行环境，其任务专注于目标条件强化学习。

This dataset is a high-performance codebase and benchmark designed for self-supervised goal-conditioned reinforcement learning (GCRL), which enables agents to conduct rapid training across diverse environments. This benchmark allows for the evaluation of key design choices in contrastive reinforcement learning, and facilitates effective testing of various energy functions and loss objectives. This dataset supports 1024 parallel environments, with its tasks focused on goal-conditioned reinforcement learning.