Towards Automated Safety Requirements Derivation Using Agent-based RAG

2025-04-15

AI Summary

To address the dual challenges of domain knowledge scarcity and imprecise retrieval under complex queries in automated safety requirement generation for autonomous driving, this paper proposes an Agent-Augmented Retrieval-Augmented Generation (Agent-RAG) framework based on multi-agent collaboration. The method integrates domain-specific knowledge graph construction, multi-hop retrieval orchestration, and real-world Apollo perception system cases to enable closed-loop optimization across query understanding, hierarchical retrieval, and result verification. It pioneers the integration of multi-agent coordination into the RAG pipeline, significantly enhancing semantic alignment and recall accuracy for critical safety documents. Evaluated on the Apollo Safety QA dataset, Agent-RAG achieves a 37% improvement in relevance score and a 52% reduction in redundant information compared to baseline RAG. Furthermore, the framework demonstrates compliance with ISO 26262 functional safety standards through rigorous validation.

Technology Category

• Natural Language Processing: Safety and Robustness
• Knowledge Representation and Reasoning: Automated Reasoning and Theorem Proving
• Multiagent Systems: Multiagent Planning

Application Category

• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Responsible Web: Machine-in-the-loop, human agency and autonomy

Abstract

We study the automated derivation of safety requirements in a self-driving vehicle use case, leveraging LLMs in combination with agent-based retrieval-augmented generation. Conventional approaches that utilise pre-trained LLMs to assist in safety analyses typically lack domain-specific knowledge. Existing RAG approaches address this issue, yet their performance deteriorates when handling complex queries and it becomes increasingly harder to retrieve the most relevant information. This is particularly relevant for safety-relevant applications. In this paper, we propose the use of agent-based RAG to derive safety requirements and show that the retrieved information is more relevant to the queries. We implement an agent-based approach on a document pool of automotive standards and the Apollo case study, as a representative example of an automated driving perception system. Our solution is tested on a data set of safety requirement questions and answers, extracted from the Apollo data. Evaluating a set of selected RAG metrics, we present and discuss advantages of a agent-based approach compared to default RAG methods.

Problem

• Automating safety requirements for self-driving vehicles using LLMs
• Improving domain-specific knowledge retrieval in safety analyses
• Enhancing query relevance in agent-based RAG for automotive standards

Innovation

• Agent-based RAG for safety requirements
• LLMs combined with domain-specific retrieval
• Automated driving perception system application