Multi agent systems improvement: evidence map - 40 findings across 40 sources

Evidence Landscape

This evidence map surveys 40 independent multi agent systems improvement sources drawn from the Tier-2 corpus and classified as direct findings. They span several populations, comparators, and endpoints and are catalogued by source in the Findings Map rather than pooled into one estimate — cross-population aggregation is not claimed. Each row records its own population, comparator, endpoint, and effect, so the spread of the literature and any tensions between findings remain explicit.

Findings Map

Population	Comparator	Finding	Source
multi agent systems accuracy tasks	isolated single-marketplace…	Our framework achieves 96.8% fraud detection accuracy with 0.31% false positive rate—a 9.1…	2026 doi:10.1109/icaic67076.2026.11395673
multi agent systems accuracy tasks	using LLM-as-Judge	AgentAuditor is agnostic to MAS setting, and we find across 5 popular settings that it yie…	2026 doi:10.48550/arxiv.2602.09341
multi agent systems accuracy tasks	traditional manual and singl…	Experiments demonstrate that compared to traditional manual and single-robot operations, t…	2026 doi:10.1088/2631-8695/ae3b9e
multi agent systems accuracy tasks	traditional optimization	This model had a high prediction and decision-making accuracy of 96.2% which is better tha…	2026 doi:10.1109/iconic67661.2026.11517785
multi agent systems accuracy tasks	strong multi-agent RL baseli…	Compared with strong multi-agent RL baselines such as Bi-AC, MACPO, and MAPPO-L, RARL achi…	2026 doi:10.4108/eetiot.10944
multi agent systems accuracy tasks	MPHunter--one of the state-o…	On D1, LAMPS achieves 97.7% accuracy, surpassing MPHunter--one of the state-of-the-art app…	2026 doi:10.1016/j.jss.2026.112792
multi agent systems accuracy tasks	settings but only 8.3% under…	Under simulated adversarial prompt injection, task accuracy declined by 29.5% in baseline…	2026 doi:10.71465/ajainn3659
multi agent systems accuracy tasks	all physician groups: pulmon…	Results NS-MAS achieved an overall accuracy of 90.0% (27/30), significantly exceeding all…	2026 doi:10.21203/rs.3.rs-9262455/v1
multi agent systems F1 tasks	strong AFE baselines	Across 15 public benchmarks (classification with macro-F1; regression with inverse relativ…	2026 doi:10.48550/arxiv.2602.16435
iterative, closed-loop designs in LLM-…	linear workflows	iterative, closed-loop designs neutralizing over 40% of faults that cause catastrophic col…	2026 doi:10.48550/arxiv.2602.19843
multi-agent systems	single-agent approaches	achieving average match improvements of 23.66% and 14.05% over single-agent and multi-agen…	2026 doi:10.48550/arxiv.2602.08335
multi agent systems recall tasks	) under instruction-data dec…	single-agent baseline) under instruction-data decoupling, and the decoupling mechanism boo…	2026 doi:10.1016/j.watres.2026.126163
multi agent systems recall tasks	the best Single-LLM (Gemini-…	The Mixed-Vendor MAC achieves a Recall@1 of 40.00%, outperforming the best Single-LLM (Gem…	2026 doi:10.18653/v1/2026.healing-1.1
multi agent systems success rate tasks	the existing approaches—with…	Experiment results demonstrated that the proposed PWS-MADDPG achieved a grasping success r…	2026 doi:10.1109/tase.2026.3672621
multi agent systems success rate tasks	vs.	However, the multi-agent system achieves a higher success rate than a single-agent system…	2026 doi:10.14429/dsj.21693
multi agent systems success rate tasks	algorithms; localization acc…	Simulation results validate the effectiveness of HMUDRL: in the later stages of training,…	2026 doi:10.3390/drones10010054
multi agent systems success rate tasks	fixed communication protocol…	Experimental results demonstrate a 25.6% improvement in task success rate and a 30.2% redu…	2026 doi:10.66238/fsrma54
multi agent systems success rate tasks	static prompt-based agents	Experimental results show that the proposed method improves task success rate from 71.3% t…	2026 doi:10.71465/ajml3665
multi agent systems success rate tasks	95.7% in the training enviro…	Velocity and spacing tracking errors are maintained within 3% and 1%, respectively, and th…	2026 doi:10.3390/electronics15091823
multi agent systems win rate tasks	(achieving a 72.13% win rate…	Results show improved performance against a next-speaker prediction baseline (achieving a…	2026 doi:10.1609/aaai.v40i48.42120
multi agent systems win rate tasks	vs.	30m), R-QMIX significantly improves both sample efficiency and final win rate (WR), for ex…	2026 doi:10.3390/robotics15010028
multi agent systems accuracy tasks	existing approaches across a…	The framework also performs strongly in detecting front running (88.9% accuracy), denial-o…	2025 doi:10.1038/s41598-025-14032-w
multi agent systems accuracy tasks	baseline methods	In experiments conducted across logistics, inspection, and search & rescue scenarios, Auto…	2025 doi:10.1109/tccn.2025.3528892
multi agent systems accuracy tasks	traditional LLM-based techni…	Rigorous experimentation shows that the approach achieves over 80% SQL generation accuracy…	2025 doi:10.1080/20964471.2025.2483541
multi agent systems accuracy tasks	the state-of-the-art solutio…	Our results demonstrate that the proposed approach reduces latency up to 44.4% while maint…	2025 doi:10.1109/tvt.2024.3520637
multi agent systems accuracy tasks	single-agent system	Our results suggest that the multi-agent system (MAS) performed better than the single-age…	2025 doi:10.1109/cibcb66090.2025.11177136
multi agent systems accuracy tasks	state-of-the-art ICP methods	Results show that the proposed ICP-MAPPO algorithm, with its dynamic-decentralized-executi…	2025 doi:10.1109/tiv.2024.3471909
multi agent systems accuracy tasks	single agents, the component…	Our results reveal a paradox: while multi-agent systems generally outperformed single agen…	2025 doi:10.48550/arxiv.2506.06574
multi agent systems accuracy tasks	an OFA baseline while mainta…	Extensive experiments on MNIST, CIFAR-10, and CIFAR-100 demonstrate that MARCO achieves a…	2025 doi:10.48550/arxiv.2506.13755
multi agent systems accuracy tasks	AI agents autonomously extra…	Overall, the framework demonstrates around a 20 % improvement in sprint planning accuracy…	2025 doi:10.1109/icwite64848.2025.11306978
multi agent systems accuracy tasks	independent learning and non…	Finally, numerical results demonstrate that the proposed algorithm, which integrates coope…	2025 doi:10.1109/vtc2025-fall65116.2025.11310364
multi agent systems accuracy tasks	baseline methods	Experimental results demonstrate superior performance compared to baseline methods, achiev…	2025 doi:10.1109/iceca66444.2025.11382981
multi agent systems accuracy tasks	traditional methods signific…	The results show that the framework achieves a daily detection accuracy of 92% and reduces…	2025 doi:10.1145/3795154.3795432
multi agent systems accuracy tasks	standalone models	The ensemble model achieved the best performance with 88.6 percent classification accuracy…	2025 doi:10.12732/ijam.v38i11s.1856
multi agent systems accuracy tasks	state-of-the-art approaches	Our comprehensive evaluation, conducted across urban, suburban, and highway scenarios with…	2025 doi:10.1109/tvt.2025.3574081
multi agent systems accuracy tasks	up to 63.15% (GPT-4o); Trial…	GPT Comparison: Extraction Accuracy: 80.29% vs up to 63.15% (GPT-4o); Trial Matching Accur…	2025 doi:10.1200/jco.2025.43.16_suppl.1554
multi agent systems accuracy tasks	traffic congestion reached 9…	The decision-making accuracy reached between 13 % and 17% improvement across various scena…	2025 doi:10.1109/icvadv63329.2025.10961787
multi agent systems accuracy tasks	Poligraph—the current state-…	Compared with Poligraph—the current state-of-the-art privacy policy analysis framework—our…	2025 doi:10.1109/aiot66900.2025.00149
multi agent systems accuracy tasks	accuracy, surpassing traditi…	For instance, at 70 percent pruning, our approach retains up to 98.23 percent of baseline…	2025 doi:10.48550/arxiv.2509.05446
multi agent systems accuracy tasks	reinforcement learning and p…	Experimental studies based on a simulated disaster recovery context demonstrate that Neuro…	2025 doi:10.5220/0014201400004932

Limitations

This is a scoping map of retrieved direct findings, not a meta-analysis: no pooled effect is computed, coverage is bounded by the Tier-2 corpus, and heterogeneity across rows precludes a single unified conclusion.

Scope

What is the range of reported effects across the multi agent systems improvement literature, and how do they vary by population, comparator, and endpoint? This map catalogues the findings rather than converging them to one claim.

Search Summary

40 direct (A_core) sources were retrieved from the Tier-2 semantic corpus for this topic and lane-classified; each is cited with a resolvable identifier in the source bundle below.

Tensions and Gaps

Findings differ in population, comparator, endpoint, and effect size, so they are not directly comparable and are not pooled. Gaps remain where a population or comparator is represented by only a single source.