ADAS Voice Response Assistant Using LLM and Vehicle Context
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing driving assistance systems struggle to quickly generate voice responses to driver questions about Advanced Driver-Assistance Systems (ADAS) functions, particularly when the vehicle is in motion.
Innovation Solution
A driving assistance device equipped with a processor that acquires, recognizes, and generates voice responses to driver questions about ADAS using Large Language Models (LLM), while considering the vehicle's state, location, and conditions for activating ADAS functions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If conventional voice recognition systems are used to answer driver questions about ADAS, then the system structure is simple, but the response speed is insufficient and cannot provide timely assistance when the vehicle is in motion
Solution Approach 1:
The patent introduces an agent server as an intermediary component that handles complex language understanding and response generation. The driving assistance device communicates with the agent server via API calls, delegating the sophisticated NLP tasks to the server while maintaining a relatively simple local device structure. This resolves the contradiction by achieving fast, intelligent responses without significantly complicating the onboard device architecture.
Solution Approach 2:
The system pre-loads and stores multiple types of information (vehicle state information, manual information, location information, ADAS function information) in advance before the driver asks questions. This preliminary preparation of data allows the LLM to generate responses quickly when queries are received, as the necessary information is already available and structured for rapid processing.
2Measurement precision
If comprehensive vehicle state information is collected to accurately answer ADAS questions, then the accuracy of voice responses is improved, but the information processing complexity increases
Solution Approach 1:
The patent segments information into distinct categories: vehicle state information (speed, acceleration, brake status), manual information (ADAS operation methods), location information (GPS coordinates, map data), and ADAS function information (function names, activation conditions). This segmentation allows the LLM to process specific types of information as needed rather than handling all data simultaneously, reducing processing complexity while maintaining comprehensive accuracy.
Solution Approach 2:
The agent server acts as an intermediary that receives structured information from the driving assistance device and performs complex information integration and query processing. This distributes the processing burden, allowing the onboard device to maintain simpler architecture while achieving high response accuracy through the server's sophisticated information handling capabilities.
3Adaptability or versatility
If the system answers all driver questions about ADAS functions, then the driver assistance capability is improved, but the time required to generate responses increases
Solution Approach 1:
The system performs preliminary actions by pre-collecting and structuring all necessary information (vehicle state, manual guidelines, location data, ADAS function details) before the driver asks questions. This advance preparation enables the LLM to generate comprehensive responses quickly when queries are received, as the information infrastructure is already in place and optimized for rapid retrieval and processing.
Solution Approach 2:
The system continuously monitors vehicle state changes and updates information accordingly, ensuring that the LLM always has current data to work with. This feedback mechanism allows the system to maintain high adaptability to different driving situations while keeping response times short, as the information is already structured and ready for immediate use rather than requiring real-time analysis.
Data Source
AI summary
A driving assistance device includes an acquisition unit which acquires a voice question about ADAS accepted from a driver driving a vehicle, a voice recognition unit which recognizes the voice question about the ADAS acquired by the acquisition unit, and a response generation unit which uses LLM and generates a voice response corresponding to the voice question about the ADAS recognized by the voice recognition unit.