Abstract
Autonomous vehicles (AVs) struggle with real-time decision-making in complex or ambiguous scenarios such as sensor conflicts, unpredictable driver behavior, or rare road events, commonly referred to as "edge cases." Current AV systems rely on pre-programmed rules, static AI models, or sensor fusion alone to interpret these events. However, these approaches often lack contextual understanding, adaptability, and transparency. As a result, AVs may respond incorrectly or fail to act altogether, leading to safety concerns and eroded public trust in autonomous technologies.
Researchers at Florida Atlantic University have developed a novel Safety Self-Talk (SST) system that uses Large Language Models (LLMs) to simulate the human reasoning process in autonomous vehicles. Unlike current solutions, which rely solely on static algorithms, SST dynamically engages in real-time analysis when sensor inconsistencies or potential risks are detected. By initiating a self-dialogue, similar to that of a human driver, the system interprets complex scenarios, recommends appropriate driving actions, and provides transparent explanations. This approach significantly enhances situational awareness, responsiveness, and passenger trust. The technology is currently at the proof-of-concept stage, with ongoing development focused on integration with AV platforms and training the LLM on real-world driving scenarios.
FAU seeks to advance this innovation into the marketplace through licensing or development partnerships.
Benefit
Contextual Reasoning - Emulates human situational awarenessImproved Safety - Enhances decision-making in unpredictable conditionsTrust Building - Offers explainability and transparency to passengersMarket Application
Autonomous Vehicles - Safer navigation and cooperationIndustrial Robotics - Context-aware machine behaviorUnmanned Aerial Systems - Smarter in-flight risk response
Brochure
