Adaptive Online Packing-guided Search for POMDPs

Authors: Chenyang Wu, Guoyu Yang, Zongzhang Zhang, Yang Yu, Dong Li, Wulong Liu, Jianye Hao

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

Reproducibility Variable Result LLM Response
Research Type Experimental We evaluate our algorithm on several tricky POMDP domains, and it outperforms the state-of-the-art in all of them. (from Abstract) and This section evaluates our method on four domains based on the POMDPs.jl framework [27] and conducts an ablation study showing the contribution of each component. (from Section 5 Introduction)
Researcher Affiliation Collaboration 1National Key Lab for Novel Software Technology, Nanjing University, Nanjing 210023, China 2Pazhou Lab, Guangzhou 510330, China 3Noah s Ark Lab, Huawei Company
Pseudocode Yes Algorithm 1 General Procedure (Section 3), Algorithm 2 PLANNING(b0) (Section 3.1), Algorithm 3 EXPAND(b) (Section 3.2), Algorithm 4 KLD-SAMPLING(b) (Section 3.2.1), Algorithm 5 PROPAGATE(b, a) (Section 3.2.1)
Open Source Code Yes Codes are available at https://github.com/LAMDA-POMDP/Ada OPS.jl.
Open Datasets Yes We evaluate our method on four domains based on the POMDPs.jl framework [27] (Section 5 Introduction) and for Roomba, [28] K. Menda, Z. Sunberg, and M. Kochenderfer, Roomba POMDPs.jl. https://github.com/sisl/Roomba POMDPs.jl, 2021.
Dataset Splits No The paper mentions that 'Hyperparameters for each algorithm are tuned via grid search', which implies a validation process. However, it does not explicitly state training/test/validation dataset splits (e.g., percentages or sample counts for a validation set) for reproducibility.
Hardware Specification Yes All experiments were conducted on a computer with Intel(R) Core(TM) i7-10750H, 6v CPUs running at 2.6GHz, and 32G main memory.
Software Dependencies No The paper states that experiments are based on the 'POMDPs.jl framework [27]', citing its publication in 2017. However, it does not provide a specific version number for POMDPs.jl or any other software dependencies like Python, PyTorch, etc.
Experiment Setup Yes Hyperparameters for each algorithm are tuned via grid search, and another independent experiment is conducted to test their performance. Please refer to Appendix D.2 for the ranges of grid search and the hyperparameters and heuristics selected for each algorithm. (Section 5.1.1 Experiment settings) and Table 2: Hyperparameter Ranges and Chosen Values (Appendix D.2).