Artificial Intelligence usage in Automotive On-Board systems

Contact us

Contactez-nous

Avant de remplir le formulaire d'inscription, veuillez lire le Privacy notice conformément à l'article 13 du règlement UE 2016/679
entrée non valable
entrée non valable
entrée non valable
entrée non valable
entrée non valable
entrée non valable
entrée non valable

Privacy

Je déclare avoir lu et pleinement compris la note d'information sur la protection des données personnelles Privacy Notice et j'exprime par la présente mon consentement au traitement de mes données personnelles par Reply SpA à des fins de marketing, en particulier pour recevoir des communications promotionnelles et commerciales ou des informations concernant des événements ou des webinaires de l'entreprise, en utilisant des moyens de contact automatisés (par exemple, SMS, MMS, fax, e-mail et applications web) ou des méthodes traditionnelles (par exemple, appels téléphoniques et courrier papier).

Artificial Intelligence (AI) is defined as the ability of a machine to reproduce human capabilities such as reasoning, planning or learning. AI in automotive has been used for decades in a number of applications such as fuel consumption, maintenance prediction or stability systems. With the rise of high-performance computer and new algorithms, AI systems are today able to process a much larger amount of data and continuously improve their decision. 3 main techniques are used:

Standard reasoning

Requires a person to program the entire code using modelling, algorithm etc. The computer is not involved in finding or improving the algorithm by itself.

Machine learning

Gives a computer the ability to learn without being programmed. This approach is based on statistics analysis that enables the computer to create an algorithm based on a set of input. The 4 most common methodologies are:



Supervised-learning: Human intervention is necessary to create the best possible algorithm. The programmer shall provide a set of data and “train” the computer to have the desired output. Example: In a computer vision system, the programmer needs to teach the computer how an object looks like by showing a set of pictures and by correcting the result until the computer can recognize the object.



Non-supervised learning: The algorithm is trained by using a set of data that have no “label”. Example: In a computer vision system, the programmer shall provide thousands to millions of inputs and the computer shall by itself classify the image and create the database of algorithm to detect objects.



Semi-supervised learning: The programmer is using a compromise from supervised and non-supervised methodologies.



Reinforced learning:: The programmer provides basic rules to an agent. The agent gives positive feedback for good action and negative feedback for wrong action. The computer shall estimate how to maximize the positive feedback. Example: In a Lane Keep Assist system, if the programmer specifies that the vehicle shall stay within the road lane, the agent gives a wrong feedback if a vehicle crosses the lane and will try to improve the algorithm.

Deep learning

Use of complex brain-like algorithms such neural network. This technique requires a very large amount of data, is longer but provides better results. Example: In a facial recognition system, the programmer shall specify the main characteristics to recognize shape, eyes, mouth, etc. and train for each characteristic. Once the characteristics are connected to each other using a neural network, the computer developed the ability to recognize faces.

Conclusion

AI usage in Automotive On-Board system will continue to increase with the improvement of computing capabilities and SW techniques. If the programmer does not need the support of the machine, a standard approach with existing algorithms/models is enough. If the programmer needs support for finding or improving the models/algorithms, machine learning or deep learning may be used.


Machine or deep learning algorithms are often considered as coming from a “Black Box”. This can create questions such as “is the computer correct”? Programmers shall be able to understand and control the way AI works and the grounds for its decisions.

Contact us
  • strip-0

    Within the Reply Group, Autonomous Reply is the specialist in the specification, development, integration and validation of autonomous and connected embedded systems. We offer a portfolio of services covering the entire value chain, from strategy definition to implementation and operational safety. Autonomous Reply's services include consulting, real-time systems engineering, software development and integration of autonomous solutions.