Improved POMDP Tree Search Planning with Prioritized Action Branching

Authors: John Mern, Anil Yildiz, Lawrence Bush, Tapan Mukerji, Mykel J. Kochenderfer11888-11894

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

Reproducibility Variable Result LLM Response
Research Type Experimental Experiments show that PA-POMCPOW is able to outperform existing state-of-the-art solvers on problems with large discrete action spaces. We tested the performance of PA-POMCPOW on two tasks involving sensor placement and wildfire containment.
Researcher Affiliation Collaboration John Mern,1 Anil Yildiz, 1 Larry Bush, 2 Tapan Mukerji, 3 and Mykel J. Kochenderfer 1 1Stanford University, Department of Aeronautics and Astronautics, 496 Lomita Mall, Stanford, CA 94305 2General Motors, Research and Development, Warren, MI 3Stanford University, Department of Energy Resources Engineering, 367 Panama Street, Stanford, CA 94305
Pseudocode Yes Algorithm 1 ACTIONPROGWIDEN Function; Algorithm 2 SELECTACTIONS Function
Open Source Code Yes The solver source code is available at https://github.com/sisl/PAPOMCPOW.jl.
Open Datasets No The paper describes how the data for the 'Sensor Placement' and 'Wildfire Containment' tasks are generated internally (e.g., 'information densities are generated by sampling from a Gaussian Process prior', 'fuel map is an array of how much fuel is contained in each cell and is generated by sampling each cell from a truncated Gaussian distribution'), but it does not provide concrete access information (link, DOI, formal citation) for a publicly available or open dataset.
Dataset Splits No The paper describes simulation settings such as '100 different initializations' and '500 simulations over 100 state and initial belief realizations', but it does not specify explicit training, validation, or test dataset splits.
Hardware Specification No The paper does not provide specific hardware details (e.g., CPU/GPU models, memory) used for running its experiments, only mentioning simulation calls and run times.
Software Dependencies No The paper mentions using 'POMDPs.jl (Egorov et al. 2017)' for implementation but does not provide specific version numbers for this or any other software dependencies required to replicate the experiments.
Experiment Setup Yes The Λ vector was set to linearly spaced values between 0 and 2 with a step size of 0.1 for a total of 20 considered actions. We ran each test with limits of 100, 500, and 1000 simulator calls per step. The Λ set was composed of linearly spaced values between 0.5 and 1.5 with a step size of 0.1 for a total of 16 values. We ran each test with limits of 100, 250, and 500 simulator calls per solver step for a grid size of 40x40.