Artificial intelligence on-board: how does it improve the in-car experience?
Digital assistants play an increasingly central role when we are in the car. How do they feed themselves? With artificial intelligence. Machine learning and ChatGPT make vehicle interaction increasingly intuitive, personal and safe.
Advanced technology
Today cars offer users a completely new experience, thanks in part to the so-called digital assistants, which accompany us in our daily lives on board. ‘The digital assistant understands and interacts, learns routines and makes suggestions proactively. The technological basis is provided by various artificial intelligence modules, which enable a symbiotic relationship between user and vehicle,’ explains Kim Cedziwoda, head of CARIAD's Digital Assistant team.
Accurate localisation of the speaker
A concrete example? Audi models with the new E3 1.2 electronic architecture, debuting in the Audi Q6 e-tron. The Audi digital assistant, developed in cooperation with CARIAD, works in self-learning mode. It is deeply integrated into the infotainment system and offers user support with more than 800 voice commands. Natural language understanding allows to control vehicle functions, make calls, configure the travel route, play music, answer questions, get information on weather or simply chat - all this is possible without taking the eyes off the road.
Deep learning models use microphonic streams to identify who is talking to the assistant, and this information is taken into account when executing commands: if the front passenger asks to open the window, the system will intervene at that position. Some functions, however, can only be activated by the driver for safety reasons.
More personalisation
The digital assistant learns user behaviour. It uses contextual information and identifies patterns in the driver's interaction with the car, such as recurring operational sequences.
Using machine learning algorithms, it distils these into routines, which it then proactively suggests the next time a similar situation occurs. For example, if the driver activates air recirculation along the route between home and work and vice versa, the system will ask him whether to activate it along the same route each time. Other examples are predictive call lists, comfort settings, climate control and more, subject to the user's consent.
Generative AI for natural dialogues
‘The goal for a voice assistant is to make dialogues as natural as possible and cover a broad spectrum of topics and information,’ Cedziwoda continues. ‘To do this, the system relies on the latest natural language processing and accesses a multitude of data sources, from vehicle data to input from online information.’
One of the new sources is ChatGPT, an artificial intelligence-based chatbot accessible via the Azure OpenAI service. This feature for Audi customers ‘relies’ on Chat Pro from Cerence Inc., a solution that allows ChatGPT to be integrated into an automotive context. And precisely the integration of ChatGPT into the automotive landscape is one of the first productive customers applications of Large Language Models (LLM).
LLMs are trained with large amounts of textual data from the Internet, books or forums. These text components are used to calculate the probabilities of word sequences with incredible precision, which improves human-computer interaction in terms of naturalness and breadth of knowledge. As an example, ChatGPT can provide information on tourist attractions, football matches already played or help solve maths problems. In the future, it will also be possible to ask questions about the vehicle, e.g. how to set the tyre pressure.
Seamless experience
All the intelligent functions of the Audi Assistant are perfectly integrated and ensure that drivers have a smooth and personalised interaction with their car. The digital assistant automatically recognises who is talking and understands whether it is necessary to perform an action on board, search for a destination or provide information such as the weather forecast.
The system only forwards queries to ChatGPT when it is unable to answer general culture questions. In all this, data security has top priority: to ensure privacy, all queries and answers are deleted after processing. Furthermore, ChatGPT never has access to vehicle data.