ML-SceGen: A Multi-level Scenario Generation Framework
Yicheng Xiao, Yangyang Sun, Yicheng Lin
TL;DR
ML-SceGen introduces a three-stage framework that combines Large Language Models (LLMs) with an Answer Set Programming (ASP) core to generate comprehensive and controllable autonomous-driving scenarios, including dangerous factors at uncontrolled intersections. The pipeline transitions from Functional to Logical to Concrete scenarios, leveraging ASP (via Clingo) for combinatorial coverage and LLMs for hazard identification and parameter tuning, with a Scenariogeneration-based output format (OpenScenario/OpenDrive). Key contributions include identifying ASP as the backbone for comprehensive scenario generation, integrating LLMs to inject risk factors, and demonstrating controllability through experiments on diverse intersection layouts. The approach promises improved realism and safety coverage for simulator-based training and validation, while outlining concrete future enhancements such as richer danger components, better Carla integration, and metric-driven evaluation.
Abstract
Current scientific research witnesses various attempts at applying Large Language Models for scenario generation but is inclined only to comprehensive or dangerous scenarios. In this paper, we seek to build a three-stage framework that not only lets users regain controllability over the generated scenarios but also generates comprehensive scenarios containing danger factors in uncontrolled intersection settings. In the first stage, LLM agents will contribute to translating the key components of the description of the expected scenarios into Functional Scenarios. For the second stage, we use Answer Set Programming (ASP) solver Clingo to help us generate comprehensive logical traffic within intersections. During the last stage, we use LLM to update relevant parameters to increase the critical level of the concrete scenario.