Asynchronous Communication Aware Multi-Agent Task Allocation
Authors: Ben Rachmut, Sofia Amador Nelke, Roie Zivan
IJCAI 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | 4 Experimental Design, 5 Experimental Evaluation, Figures 1 and 2 present the team utility of the allocations generated by the different algorithms, as a function of NCLOs, when solving static problems. |
| Researcher Affiliation | Academia | 1Ben Gurion University of the Negev 2Holon Institute of Technology |
| Pseudocode | Yes | Algorithm 1 FMC ATA code of Task Agent taj, Algorithm 2 FMC ATA code of Active Agent aai |
| Open Source Code | Yes | The simulator s code is public and available1 (with footnote 1: https://github.com/benrachmut/Simulation For Research) |
| Open Datasets | No | In each experiment, we randomly generated 50 different problem instances. The paper does not provide concrete access information (link, DOI, specific citation) for a publicly available dataset, instead describing a process for generating problem instances. |
| Dataset Splits | No | The paper does not explicitly provide training/validation/test dataset splits. It describes generating '50 different problem instances' for each experiment run. |
| Hardware Specification | No | The paper states 'agents were implemented as Python threads' and refers to a 'distributed asynchronous simulator', but no specific hardware details (e.g., CPU/GPU models, memory) used for the experiments are provided. |
| Software Dependencies | No | The paper mentions 'agents were implemented as Python threads' but does not specify any particular software dependencies with version numbers (e.g., libraries, frameworks). |
| Experiment Setup | Yes | In each experiment, we randomly generated 50 different problem instances. Each scenario in an experiment included two types of agents, active agents and task agents, and for each of them, a random geographic location (coordinates x and y) was selected uniformly between 0 and 10^6. Each problem instance included active agents with a set of three unique skills. An active agent was characterized by its set of skills. The skills for each active agent were selected randomly with probability 0.5 for possessing each skill. If the result of the process was that the agent did not possess any skill, one was selected randomly and assigned to it. All agents the same consistent speed, a single unit of distance per NCLO. Each task requires three skills. The value of nk j (i.e., maximum number of agents required for handling vk j ) was consistent and set to 5. For each skill in a task, the value of Cap(vk j , nk j ) was selected from a uniform distribution as follows: Cap(vk j , nk j ) U(0, 10^5). For q < nk j the value was relative such that Cap(vk j , q) = Cap(vk j , nk j ) q The workload required (wk j ) was also selected from a uniform distribution and was affected by Cap(vk j , nk j ) such that wk j U(10^5, 10^5 + Cap(vk j , nk j )). In Algorithm 1: pk p| pk < ϵ (in our experiments we used ϵ = 10^-5). The time between the addition of new arriving tasks (tbt) was randomly selected from an exponential distribution, i.e., tbt exp(β) where the parameter β controls the mean time between tasks. We used β = 10^5 in our experiments. |