Multimodal Atari Games

该数据集专为深度强化学习代理设计，包含了多种模态的场景，旨在评估在感知信息不完整的情况下的性能表现。具体包括两种场景：摆杆和超热点，它们都涉及到模态信息不完整所带来的特定挑战。数据集的评估采用了100个剧集和10次随机种子运行。任务集中在多模态场景下的强化学习。

This dataset is purpose-built for deep reinforcement learning (DRL) agents, covering a range of multimodal scenarios with the core goal of evaluating agent performance under incomplete perceptual information. It includes two specific scenarios: CartPole and Hyper Hotspot, both of which present distinct challenges arising from incomplete modal sensory information. The dataset's evaluation setup uses 100 episodes per run and 10 runs with distinct random seeds. The tasks focus on reinforcement learning in multimodal scenarios.