Safety-constrained and runtime-assured LLM multi-agent decision making for intelligent vehicle systems
- 1Beijing Information Technology College, Beijing, China.
Abstract
Large language models offer flexible semantic reasoning for intelligent-vehicle interfaces, but probabilistic outputs, multi-agent disagreement, and provider failures complicate safety-sensitive use. We developed SCI-Proposed-v1.0, a lightweight multi-agent semantic decision architecture that combines transparent risk-informed routing with an independent Safety Validator and deterministic local Runtime Fallback. The system produces canonical semantic actions rather than trajectories or direct vehicle-control commands. In a frozen confirmatory benchmark of 150 unseen Base Scenarios, each represented by three reviewed language variants, SCI-Proposed completed 360/450 inputs (80.0%), compared with 169/450 (37.6%) for EI-MAS and 155/450 (34.4%) for a Single LLM. Language variants were treated as repeated observations within each Base Scenario, and all methods used the same frozen model configuration. Base-scenario paired differences were 0.4244 (95% confidence interval, 0.3489–0.5000; p < 0.001) and 0.4556 (95% confidence interval, 0.3800–0.5311; p < 0.001), respectively. Complete Safety Requirement Satisfaction was 316/360 (87.8%) versus 101/360 (28.1%) for EI-MAS. In frozen ablation, removing the Safety Validator reduced its prespecified endpoint by 0.2767 (p < 0.001), and removing Runtime Fallback reduced represented-failure recovery by 0.3667 (p = 0.00195). Exact three-of-three canonical action-set stability was 27/30 (90.0%). In prespecified isolated-sandbox hardware-in-the-loop validation, Scientific Task Success was 25/30, Safety Enforcement Success was 7/7 applicable cases, and Runtime Recovery was 4/6 applicable cases. Across these complementary, prespecified stages, separating generative reasoning from independent safety validation and class-bounded local runtime recovery materially improved semantic-decision dependability under the tested qwen-plus/DashScope configuration, frozen semantic policy, and represented fault classes. These findings do not establish broader vehicle-system safety, physical-control reliability, arbitrary-fault recovery, or deployment readiness.
Related
- A hierarchical navigation decision-making method for UAV swarms in unknown communication-constrained environments
- Epigenetic nutrition in aquaculture: The emerging role of Artemia as a bioencapsulation vehicle
- An embedded deep learning–fuzzy logic framework for intelligent and automated irrigation in smart farming
- Multi-scale lodging in rice and wheat: bridging biomechanical traits, phenotypic detection, and intelligent harvesting systems
- GEO-BERT: A Retrieval augmented LLM Framework for Intelligent Dataset Retrieval from the Gene Expression Omnibus
BibTeX
@article{dang2026safetyconstrained,
title = {{Safety-constrained and runtime-assured LLM multi-agent decision making for intelligent vehicle systems}},
author = {Wei Dang and Lijun Gao},
journal = {PLOS One},
year = {2026},
month = oct,
doi = {10.1371/journal.pone.0357928},
url = {https://doi.org/10.1371/journal.pone.0357928},
note = {CC BY},
}
RIS
TY - JOUR
TI - Safety-constrained and runtime-assured LLM multi-agent decision making for intelligent vehicle systems
AU - Wei Dang
AU - Lijun Gao
PY - 2026
DA - 2026/10/30
JO - PLOS One
DO - 10.1371/journal.pone.0357928
UR - https://doi.org/10.1371/journal.pone.0357928
AB - Large language models offer flexible semantic reasoning for intelligent-vehicle interfaces, but probabilistic outputs, multi-agent disagreement, and provider failures complicate safety-sensitive use. We developed SCI-Proposed-v1.0, a lightweight multi-agent semantic decision architecture that combines transparent risk-informed routing with an independent Safety Validator and deterministic local Runtime Fallback. The system produces canonical semantic actions rather than trajectories or direct vehicle-control commands. In a frozen confirmatory benchmark of 150 unseen Base Scenarios, each represented by three reviewed language variants, SCI-Proposed completed 360/450 inputs (80.0%), compared with 169/450 (37.6%) for EI-MAS and 155/450 (34.4%) for a Single LLM. Language variants were treated as repeated observations within each Base Scenario, and all methods used the same frozen model configuration. Base-scenario paired differences were 0.4244 (95% confidence interval, 0.3489–0.5000; p < 0.001) and 0.4556 (95% confidence interval, 0.3800–0.5311; p < 0.001), respectively. Complete Safety Requirement Satisfaction was 316/360 (87.8%) versus 101/360 (28.1%) for EI-MAS. In frozen ablation, removing the Safety Validator reduced its prespecified endpoint by 0.2767 (p < 0.001), and removing Runtime Fallback reduced represented-failure recovery by 0.3667 (p = 0.00195). Exact three-of-three canonical action-set stability was 27/30 (90.0%). In prespecified isolated-sandbox hardware-in-the-loop validation, Scientific Task Success was 25/30, Safety Enforcement Success was 7/7 applicable cases, and Runtime Recovery was 4/6 applicable cases. Across these complementary, prespecified stages, separating generative reasoning from independent safety validation and class-bounded local runtime recovery materially improved semantic-decision dependability under the tested qwen-plus/DashScope configuration, frozen semantic policy, and represented fault classes. These findings do not establish broader vehicle-system safety, physical-control reliability, arbitrary-fault recovery, or deployment readiness.
SN - 1932-6203
ER -