The procedure followed while creating this data is summarized in Section II of Chen, Brian, et al. "Behavioral cloning in atari games using a combined variational autoencoder and predictor model." 2021 IEEE Congress on Evolutionary Computation (CEC). IEEE, 2021. This data is not a result of a research but an intermediate product that is used in research.
This dataset is generated to train a behavioral cloning framework from gameplay screen captures and keystrokes of an "expert" player. The RL agent that is trained using "RL Baselines Zoo package" acts as the "expert" player, whose decision making process we desire to learn. In addition to behavioral cloning experiments, this dataset is further used to demonstrate the efficacy of a novel incremental tensor decomposition algorithm on image-based data streams.
Chen, Brian, et al. "Behavioral cloning in atari games using a combined variational autoencoder and predictor model." 2021 IEEE Congress on Evolutionary Computation (CEC). IEEE, 2021., Aksoy, Doruk, et al. "An Incremental Tensor Train Decomposition Algorithm." arXiv preprint arXiv:2211.12487 (2022)., and Chen, Brian, et al. "Low-Rank Tensor-Network Encodings for Video-to-Action Behavioral Cloning", forthcoming
