Presentation
From Code to Collaboration: A Cognitive Agent Framework for Large Language Model (LLM)-Based Human-Vehicle Teaming
SessionWednesday Poster Session
DescriptionAs automated vehicles become more advanced, safe use will depend not only on vehicle performance, but also on how well people understand, trust, and coordinate with automated systems. However, current research on LLM-based driving systems remains fragmented, and there is no unified framework explaining how LLMs can be integrated into automated vehicles to support human-vehicle teaming. This study examines how large language models (LLMs), such as conversational artificial intelligence (AI) systems, can support human-vehicle collaboration. A systematic review of recent studies on LLM-based systems in driving contexts was conducted, and prior research was categorized into a four-capability framework: perception and awareness, reasoning and decision-making, action and control, and interaction and communication. Findings show that LLMs can help vehicles describe driving scenes, reason about safety risks, explain decisions, support takeover communication, and improve shared situational awareness. However, risks such as hallucinations, weak physical grounding, unpredictable reasoning, and real-time delays suggest that LLMs should not be treated as standalone vehicle controllers. Instead, they should serve as high-level cognitive partners that coordinate with verified safety-critical vehicle modules. This review highlights the need for human-centered, safety-focused LLM driving systems that support transparency, calibrated trust, authority sharing, and collaboration between human drivers and automated vehicles.
Event Type
Poster
TimeWednesday, October 21st5:30pm - 6:30pm PDT
Location
Similar Presentations