AI-based embedded systems for autonomous driving

The transportation industry (automotive, railway and avionics) continues to look for ways to reduce the fatalities and the severity of accidents. Autonomous driving (AD) not only reduces the number of accidents, but offers also a better use of road infrastructures and may protect the environment. However, AD comes with inherent challenges. Specifically, many of the actions taken by the autonomous vehicle are based on increasingly complex algorithms, mainly applied from the artificial intelligence (AI) domain such as deep neural networks (DNN). These algorithms are known for their greed of computing and memory resources.In this presentation, I will talk about projects we are developing at Université Polytechnique Hauts-de-France in the design of optimized embedded systems for highly complex AD functionalities. The use of techniques such approximate computing, dynamic and partial reconfiguration and hierarchical cloud/fog/edge platforms will be explored.