Dec-SGTS: Decentralized Sub-Goal Tree Search for Multi-Agent Coordination
Authors: Minglong Li, Zhongxuan Cai, Wenjing Yang, Lixia Wu, Yinghui Xu, Ji Wang11282-11289
AAAI 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We conduct the experiments on courier dispatching problem, and the results show that Dec-SGTS achieves much better reward while enjoying a significant reduction of planning time and communication cost compared with Dec-MCTS (Decentralized Monte Carlo Tree Search). |
| Researcher Affiliation | Collaboration | Minglong Li, 1 Zhongxuan Cai, 1 Wenjing Yang, 1 Lixia Wu,2 Yinghui Xu,2 Ji Wang1 1 Institute for Quantum Information & State Key Laboratory of High Performance Computing, College of Computer, National University of Defense Technology, China 2 Artificial Intelligence Department, Zhejiang Cainiao Supply Chain Management Co., Ltd., China |
| Pseudocode | Yes | Algorithm 1 Dec-SGTS for agent i" and "Algorithm 2 Expansion with Subgoal States |
| Open Source Code | Yes | For formalization and proof, see supplementary material https://github.com/HPCL-micros/dec-sgts. |
| Open Datasets | No | The paper mentions using "CDP benchmark" and shows examples from "Amap" and "CDP grid world" but does not provide a direct link, DOI, specific repository name, or formal citation for accessing the dataset used in experiments. |
| Dataset Splits | No | No specific dataset split information (exact percentages, sample counts, citations to predefined splits, or detailed splitting methodology) is provided. |
| Hardware Specification | Yes | Dec-SGTS is implemented with a platform of 12 cores, 3.7 GHz and 16 GB Memory. |
| Software Dependencies | No | No specific ancillary software details, such as library or solver names with version numbers, are provided. |
| Experiment Setup | Yes | Agent moves for one grid with reward -0.01 and picks up a package with reward 1.0. For each experiment, we use different settings and parameters. Given planning time of 15 seconds. We use the action coverage threshold σ to expand tree nodes with subgoal states. |