Asymmetric Action Abstractions for Planning in Real-Time Strategy Games
Authors: Rubens O. Moraes, Mario A. Nascimento, Levi H.S. Lelis
JAIR 2022 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | An extensive set of experiments in a real-time strategy game shows that search algorithms using asymmetric abstractions are able to outperform all other search algorithms tested. |
| Researcher Affiliation | Academia | Rubens O. Moraes EMAIL Departamento de Inform atica, Universidade Federal de Vi cosa, Brazil Mario A. Nascimento EMAIL Department of Computing Science University of Alberta, Canada Levi H. S. Lelis EMAIL Alberta Machine Intelligence Institute (Amii) Department of Computing Science University of Alberta, Canada |
| Pseudocode | Yes | Algorithm 1 Alpha-Beta Algorithm 2 Na ıve MCTS Algorithm 3 Select And Expand Node |
| Open Source Code | No | The paper states: "Finally, the µRTS codebase1 contains most of the current state-of-the-art search algorithms, including the systems used in µRTS competitions (Onta n on, Barriga, Silva, Moraes, & Lelis, 2018), thus facilitating our empirical evaluation. 1. https://github.com/santiontanon/microrts". This refers to the open-source game environment that the authors used for their experiments, not the source code for their specific methodologies (GAB, SAB, A3N) described in the paper. |
| Open Datasets | Yes | We evaluate our algorithms in µRTS (Onta n on, 2013), an RTS game developed for research purposes. µRTS is a great testbed for our research because it offers an efficient forward model of the game, which is required by search-based approaches. Moreover, the game is much simpler than commercial video games, which allows us to evaluate different algorithms without the technical difficulties typical of commercial video games. Finally, the µRTS codebase1 contains most of the current state-of-the-art search algorithms, including the systems used in µRTS competitions (Onta n on, Barriga, Silva, Moraes, & Lelis, 2018), thus facilitating our empirical evaluation. 1. https://github.com/santiontanon/microrts |
| Dataset Splits | No | The paper describes experimental protocols for evaluating agents within the µRTS game environment, such as the number of matches played per map and how starting locations are alternated. However, it does not specify traditional training/validation/test splits for a fixed dataset, as is common in supervised machine learning contexts. It describes evaluation scenarios rather than dataset splits. |
| Hardware Specification | No | The paper does not provide specific details about the hardware (e.g., CPU, GPU models, memory) used to run the experiments. It only mentions that search-based approaches "can be run over commodity computing devices" without specifying what was actually used. |
| Software Dependencies | No | The paper mentions using the µRTS codebase but does not provide specific version numbers for any programming languages, libraries, or other software dependencies beyond the game environment itself. |
| Experiment Setup | Yes | Each player is allowed 100 milliseconds for planning in each decision-point of the game. Every match in our experiments is limited by a number of game cycles... We use 10 maps of varied sizes in our experiments... Each tested algorithm plays against every other algorithm ten times in each map... The evaluation function Ψ used with our algorithms is the average result of two random play-outs... We selected n = 200 for PGS, SSS, GAB, and SAB and n = 100 for A1N, A2N, and A3N. The set of scripts used by PGS, SSS, GAB, SAB, and A1N is worker rush (WR), light rush (LR), heavy rush (HR), and ranged rush (RR)... The ABCD algorithm used in GAB and SAB uses the technique called scripted move ordering... We use the following set of type systems, which are ordered from the coarsest (fewer types) to the finest type system (more types): {(T0), (attack type), (attack type, mobility), (attack type, mobility, hp(l = 1)), (attack type, mobility, hp(l = 2)), (attack type, mobility, hp(l = 3))}. |