Research

Paper

AI LLM February 27, 2026

SafeGen-LLM: Enhancing Safety Generalization in Task Planning for Robotic Systems

Authors

Jialiang Fan, Weizhe Xu, Mengyu Liu, Oleg Sokolsky, Insup Lee, Fangxin Kong

Abstract

Safety-critical task planning in robotic systems remains challenging: classical planners suffer from poor scalability, Reinforcement Learning (RL)-based methods generalize poorly, and base Large Language Models (LLMs) cannot guarantee safety. To address this gap, we propose safety-generalizable large language models, named SafeGen-LLM. SafeGen-LLM can not only enhance the safety satisfaction of task plans but also generalize well to novel safety properties in various domains. We first construct a multi-domain Planning Domain Definition Language 3 (PDDL3) benchmark with explicit safety constraints. Then, we introduce a two-stage post-training framework: Supervised Fine-Tuning (SFT) on a constraint-compliant planning dataset to learn planning syntax and semantics, and Group Relative Policy Optimization (GRPO) guided by fine-grained reward machines derived from formal verification to enforce safety alignment and by curriculum learning to better handle complex tasks. Extensive experiments show that SafeGen-LLM achieves strong safety generalization and outperforms frontier proprietary baselines across multi-domain planning tasks and multiple input formats (e.g., PDDLs and natural language).

Metadata

arXiv ID: 2602.24235
Provider: ARXIV
Primary Category: cs.RO
Published: 2026-02-27
Fetched: 2026-03-02 06:04

Related papers

Raw Data (Debug)
{
  "raw_xml": "<entry>\n    <id>http://arxiv.org/abs/2602.24235v1</id>\n    <title>SafeGen-LLM: Enhancing Safety Generalization in Task Planning for Robotic Systems</title>\n    <updated>2026-02-27T18:06:10Z</updated>\n    <link href='https://arxiv.org/abs/2602.24235v1' rel='alternate' type='text/html'/>\n    <link href='https://arxiv.org/pdf/2602.24235v1' rel='related' title='pdf' type='application/pdf'/>\n    <summary>Safety-critical task planning in robotic systems remains challenging: classical planners suffer from poor scalability, Reinforcement Learning (RL)-based methods generalize poorly, and base Large Language Models (LLMs) cannot guarantee safety. To address this gap, we propose safety-generalizable large language models, named SafeGen-LLM. SafeGen-LLM can not only enhance the safety satisfaction of task plans but also generalize well to novel safety properties in various domains. We first construct a multi-domain Planning Domain Definition Language 3 (PDDL3) benchmark with explicit safety constraints. Then, we introduce a two-stage post-training framework: Supervised Fine-Tuning (SFT) on a constraint-compliant planning dataset to learn planning syntax and semantics, and Group Relative Policy Optimization (GRPO) guided by fine-grained reward machines derived from formal verification to enforce safety alignment and by curriculum learning to better handle complex tasks. Extensive experiments show that SafeGen-LLM achieves strong safety generalization and outperforms frontier proprietary baselines across multi-domain planning tasks and multiple input formats (e.g., PDDLs and natural language).</summary>\n    <category scheme='http://arxiv.org/schemas/atom' term='cs.RO'/>\n    <category scheme='http://arxiv.org/schemas/atom' term='cs.AI'/>\n    <published>2026-02-27T18:06:10Z</published>\n    <arxiv:comment>12 pages, 6 figures</arxiv:comment>\n    <arxiv:primary_category term='cs.RO'/>\n    <author>\n      <name>Jialiang Fan</name>\n    </author>\n    <author>\n      <name>Weizhe Xu</name>\n    </author>\n    <author>\n      <name>Mengyu Liu</name>\n    </author>\n    <author>\n      <name>Oleg Sokolsky</name>\n    </author>\n    <author>\n      <name>Insup Lee</name>\n    </author>\n    <author>\n      <name>Fangxin Kong</name>\n    </author>\n  </entry>"
}