Hyundai Motor Group’s global software center, 42dot, has unveiled the core technology of its large language model (LLM) in-vehicle voice AI agent, Gleo AI. The system debuted this month in Hyundai Motor Company’s updated sedan, The New Grandeur, marking a shift from basic command recognition to conversational vehicle control.
Unlike conventional voice assistants restricted to fixed phrases, Gleo AI allows occupants to interact using natural speech. The system processes context, delivers real-time information, and directly manages vehicle interior functions without requiring physical button or screen inputs.
Contextual Intent and Execution Architecture
The system incorporates spatial and situational awareness to process user intent. It identifies the speaker's seating position to apply targeted adjustments and evaluates the conversational context alongside real-time driving conditions.
Once user intent is verified, Gleo AI automatically selects and coordinates the appropriate sub-agents to execute tasks, such as updating navigation destinations or modifying climate control settings.
Technically, 42dot has integrated the entire process into a unified cognition-judgment-execution workflow. The architecture routes tasks dynamically across multiple LLMs depending on the specific objective, balancing computing efficiency with operational flexibility.
Hybrid AI and Data Governance
To eliminate latency and maintain system stability, 42dot implemented a hybrid AI framework:
On-Device Processing: Time-sensitive system tasks and direct vehicle controls are handled locally inside the vehicle to guarantee low-latency performance.
Cloud Infrastructure: Operations requiring extensive computational capacity utilize cloud servers to maintain quick response speeds.
Real-time data acquisition is managed via a dedicated knowledge agent. This component performs live web searches and references proprietary datasets to summarize current information not included in the model's baseline training. The framework also supports integration with third-party external applications.
Safety and data privacy are controlled through an in-house developed Guardrail Agent. This security layer actively monitors for and limits responses to dangerous utterances, inappropriate requests, or instructions that conflict with traffic regulations. For physical vehicle operations, the system verifies the current driving safety status before executing the mechanical command. Proprietary data governance protocols secure the entire pipeline from collection to storage.
Future Implementation and Roadmap
Moving forward, 42dot will utilize Over-The-Air (OTA) wireless updates to integrate actual driving data and user feedback directly into the service.
The mid-to-long-term development strategy includes increasing the utilization of 42dot’s proprietary LLMs and expanding on-device AI capabilities. The project aims to transition the system into a predictive, personalized AI that analyzes passenger behavioral patterns and preferences to customize services proactively.
Comments
No comments yet. Be the first.