South Korean researchers are partnering with automakers to develop and commercialize next-generation end-to-end (E2E) autonomous driving technology, where artificial intelligence learns driving strategies from real-world road data.
The Electronics and Telecommunications Research Institute (ETRI) announced Thursday it will collaborate with automakers to develop AI-based autonomous driving technology that integrates learning across the entire process—from sensor recognition to driving decisions and vehicle control—using large-scale driving data collected from actual roads. The institute aims to commercialize the technology for application in real vehicles.
The technology will enable AI to comprehensively understand road environments in a manner similar to humans, implementing what researchers call "Driving Intelligence" that autonomously controls steering, acceleration, and deceleration. The goal is to scale AI software technology secured through national R&D to industrial application levels.
Conventional autonomous driving technology typically processes perception, decision-making, and control functions sequentially through separate systems. This research develops an E2E approach where a single integrated AI model understands road conditions and simultaneously determines steering, acceleration, and deceleration.
The research is expected to advance into applying multimodal models—a recent trend in global AI technology—to autonomous driving. Multimodal models are AI architectures designed to make more sophisticated decisions by understanding not only visual information from cameras but also linguistic concepts and situational context.
Through this approach, researchers plan to implement next-generation autonomous driving AI capable of human-like judgment even at complex intersections and in diverse road environments.
ETRI researchers explained that this study represents an important case of advancing core AI technology—secured through national R&D projects by the Ministry of Science and ICT, the Institute for Information & Communications Technology Planning & Evaluation (IITP), and the Korea Autonomous Driving Innovation Foundation (KADIF)—to industrial levels by combining it with real vehicle data.
The research incorporates ETRI's reinforcement learning-based autonomous driving AI software technology that predicts dangerous situations and determines optimal driving behavior in congested road environments, as well as core perception and decision-making AI software technology enabling stable driving even in adverse weather or unpredictable road conditions.
Researchers are also pursuing a new approach that combines camera-based visual information with AI's logical reasoning capabilities, moving away from heavy reliance on expensive LiDAR sensors. This will enable development of an intelligent driving model capable of stable autonomous driving with minimal sensor configurations.
The team plans to continuously advance driving intelligence by utilizing large-scale driving data and vehicle movement information collected from actual roads for AI training. The ultimate goal is to develop this into a next-generation autonomous driving AI foundation model applicable to various vehicles and environments.
In the global market, Tesla is expanding its E2E autonomous driving software (FSD), while Nvidia has announced its own autonomous driving AI foundation model, intensifying technological competition. This research holds significant meaning as it aims to enhance global competitiveness in autonomous driving technology through close collaboration between automakers and national research institutions.
Under this partnership, automaker KG Mobility will provide real vehicle driving data and testing infrastructure, while ETRI and autonomous driving specialist Sodis will handle next-generation E2E autonomous driving software development and integrated system construction. The three organizations plan to collaborate throughout the entire process—from technology development and demonstration to industrial application—to maximize technological synergy.
"Automobiles are advanced AI systems that must simultaneously perform perception, decision-making, and control in complex road environments," said Choi Jeong-dan, head of ETRI's AI & Robotics Research Division. "Based on AI technology secured through national R&D and large-scale real driving data, we will establish guidelines for E2E autonomous driving data construction and utilization, contributing to creating an autonomous driving data ecosystem that the entire industry can leverage."
Researchers plan to proceed in phases with data collection from real road environments, AI model training, and actual vehicle application demonstrations. The ultimate goal is to develop this technology into general-purpose mobility intelligence AI applicable not only to automobiles but also to robots, drones, and various other mobile platforms.