Sequential Generative Exploration Model for Partially Observable Reinforcement Learning

Authors: Haiyan Yin, Jianda Chen, Sinno Jialin Pan, Sebastian Tschiatschek10700-10708

AAAI 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental To evaluate our method, we conduct extensive experiments on challenging 3D navigation tasks in Vi ZDoom and Deep Mind Lab. Empirical evaluation results show that our proposed exploration method could lead to significantly faster convergence than various state-of-the-art exploration approaches in the testified navigation domains.
Researcher Affiliation Academia Haiyan Yin1, Jianda Chen1, Sinno Jialin Pan1, Sebastian Tschiatschek2 1School of Computer Science and Engineering, Nanyang Technological University, Singapore 2Faculty of Computer Science, University of Vienna, Vienna, Austria
Pseudocode Yes Algorithm 1 Policy Training with SGEM
Open Source Code Yes Source code for SGEM is available in tensorflow where the details on implementation and hyperparameter settings for each task domain are also available.
Open Datasets Yes Doom My Way Home-v0 from Vi ZDoom (Kempka et al. 2016); 2) Stairway to Melon from Deep Mind Lab (Beattie et al. 2016); 3) Explore Goal Locations from Deep Mind Lab.
Dataset Splits No The paper discusses training and learning curves but does not provide specific dataset split information (exact percentages, sample counts, or citations to predefined splits) needed to reproduce the data partitioning into training, validation, and test sets.
Hardware Specification No The paper does not provide specific hardware details (e.g., exact GPU/CPU models, processor types, or memory amounts) used for running its experiments.
Software Dependencies No The paper mentions 'Source code for SGEM is available in tensorflow' but does not provide specific version numbers for TensorFlow or any other software dependencies, which are necessary for reproducible descriptions.
Experiment Setup Yes SGEM adopts action sequence length of 6 for the Vi ZDoom tasks and 3 for Deep Mind Lab tasks.