Generating Out-Of-Distribution Scenarios Using Language Models

TL;DR

This paper uses LLM strengths to introduce a framework for generating diverse OOD driving scenarios, and introduces a new “OOD-ness” metric, which quantifies how much the generated scenarios deviate from typical urban driving conditions.

Abstract

The deployment of autonomous vehicles controlled by machine learning techniques requires extensive testing in diverse real-world environments, robust handling of edge cases and out-of-distribution scenarios, and comprehensive safety validation to ensure that these systems can navigate safely and effectively under unpredictable conditions. Addressing Out-Of-Distribution (OOD) driving scenarios is essential for enhancing safety, as OOD scenarios help validate the reliability of the models within the vehicle's autonomy stack. However, generating OOD scenarios is challenging due to their long-tailed distribution and rarity in urban driving dataset. Recently, Large Language Models (LLMs) have shown promise in autonomous driving, particularly for their zero-shot generalization and common-sense reasoning capabilities. In this paper, we leverage these LLM strengths to introduce a framework for generating diverse OOD driving scenarios. Our approach uses LLMs to construct a branching tree, where each branch represents a unique OOD scenario. These scenarios are then simulated in the CARLA simulator using an automated framework that aligns scene augmentation with the corresponding textual descriptions. We evaluate our framework through extensive simulations, and assess its performance via a diversity metric that measures the richness of the scenarios. Additionally, we introduce a new "OOD-ness" metric, which quantifies how much the generated scenarios deviate from typical urban driving conditions. Furthermore, we explore the capacity of modern Vision-Language Models (VLMs) to interpret and safely navigate through the simulated OOD scenarios. Our findings offer valuable insights into the reliability of language models in addressing OOD scenarios within the context of urban driving.

Figures (10)

• Figure 2: Snapshots of OOD scenarios implemented in the CARLA simulator dosovitskiy2017carla. In both instances, the ego vehicle, represented by the red Audi, encounters challenging situations: (a) navigating through dense fog with restricted visibility, and (b) driving on a highway when a nearby vehicle abruptly switches into the ego's lane.
• Figure 3: Illustration of employing an LLM in a few-shot CoT approach to construct an initial tree for generating OOD scenarios.
• Figure 4: Employing the red-LLM to enrich the diversity of the initial tree, which leads to the diverse tree.
• Figure 5: Pruning the diverse tree to a simulatable tree, based on the available assets in CARLA.
• Figure 6: Automating the simulation of OOD scenarios in CARLA. The tree-LLM employs the simulatable tree to generate the textual description of an OOD scenario. The augmenter-LLM leverages CARLA assets to determine a scene configuration that aligns with the description of the OOD scenario.

Theorems & Definitions (1)

• Remark 1