GARLIC: GPT-Augmented Reinforcement Learning with Intelligent Control for Vehicle Dispatching
Authors: Xiao Han, Zijian Zhang, Xiangyu Zhao, Yuanshao Zhu, Guojiang Shen, Xiangjie Kong, Xuetao Wei, Liqiang Nie, Jieping Ye
AAAI 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experiments conducted on two real-world datasets demonstrate that GARLIC effectively aligns with driver behaviors while reducing the empty load rate of vehicles. Extensive experiments on two real-world road networks against advancing baselines demonstrate the effectiveness and efficiency of GARLIC. |
| Researcher Affiliation | Collaboration | 1City University of Hong Kong 2Jilin University 3Zhejiang University of Technology 4Southern University of Science and Technology 5Harbin Institute of Technology (Shenzhen) 6Alibaba Group |
| Pseudocode | No | The paper describes the methodology using mathematical formulations and descriptive text, but it does not include a clearly labeled 'Pseudocode' or 'Algorithm' block. |
| Open Source Code | Yes | Code https://github.com/Applied-Machine-Learning Lab/GARLIC |
| Open Datasets | Yes | Dataset. We use two datasets with different scales for experiments: one is located in lower and midtown Manhattan, New York City, USA2, and the other larger dataset is the taxi trajectory data from the core area of Hangzhou, Zhejiang Province, CHN3. More details can be found in Appendix A. 2https://data.cityofnewyork.us/Transportation/2018-Yellow Taxi-Trip-Data/t29m-gskq/about data |
| Dataset Splits | No | The paper mentions using two datasets for experiments and discusses the number of epochs and runs for evaluation, but it does not provide specific percentages or counts for training, validation, and test splits within the main text. |
| Hardware Specification | No | The paper describes communication aspects like V2V and network congestion but does not provide specific hardware details (e.g., GPU/CPU models, memory) used for running the experiments or training the models. |
| Software Dependencies | No | The paper mentions 'Kafuka engine' and 'cloud-edge collaboration technologies' as future work, but it does not specify any software dependencies with version numbers for the current implementation. |
| Experiment Setup | Yes | Implementation Detail To avoid network congestion, we only allow V2V communications between vehicles in adjacent regions. In addition, we limit the waiting time for vehicles to broadcast and receive V2V multi-hop messages across different regions to 1 second, ignoring any timed-out transmissions. The performance of all methods is the average result of the last 100 epochs in a total of 1000 runs. Therefore, in this paper, we set α = 0.67 to let the dispatch strategy increase the driver s income as much as possible while satisfying each driver s driving behavior. |