Episodic Future Thinking Mechanism for Multi-agent Reinforcement Learning

Authors: Dongsu Lee, Minhae Kwon

NeurIPS 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental To evaluate the proposed mechanism, we consider the multi-agent autonomous driving scenario with diverse driving traits and multiple particle environments. Simulation results demonstrate that the EFT mechanism with accurate character inference leads to a higher reward than existing multi-agent solutions.
Researcher Affiliation Academia Dongsu Lee and Minhae Kwon Department of Intelligent Semiconductors School of Electronic Engineering Soongsil University, Seoul, South Korea
Pseudocode Yes Algorithm 1 Multi-character policy training; Algorithm 2 Character inference module; Algorithm 3 Episodic future thinking mechanism
Open Source Code No Project Web: https://sites.google.com/view/eftm-neurips2024.
Open Datasets Yes To implement this task, we use the FLOW framework [58, 21, 8]. The scenario includes multiple automated vehicles on the highway. ... To confirm the scalability of the proposed solution, we also provide simulation results with a multiple particle environment (MPE) [29] and starcraft multi-agent challenge (SMAC) [48], a popular MARL testbed.
Dataset Splits No The paper mentions training and testing but does not explicitly provide percentages or counts for training, validation, or test dataset splits.
Hardware Specification Yes Appendix B System Specification CPU AMD Ryzen 9 3950X 16-core GPU Ge Forece RTX 2080 Ti RAM 128 GB SSD 1T
Software Dependencies No Appendix C.1 and C.2 list hyperparameters for Algorithm 1 and Algorithm 2 but do not specify software dependencies (e.g., programming languages, libraries, frameworks) with version numbers.
Experiment Setup Yes Appendix C Hyperparameters C.1 Algorithm 1 [list of hyperparameters and values]; C.2 Algorithm 2 [list of hyperparameters and values]