The development of Edge AI in critical and safety-critical systems raises the challenge of mastering the whole conception and implementation chain of learning components. Currently, the backend part of the state-of-the-art implementation frameworks for edge platforms are mainly "black boxes", and leads to a "black box inference engine". This engine is difficult to master and to characterize (memory allocation and transfer, computation allocations, execution time, resource sharing). This black box should be open to allow the deployment of critical AI.The objectives of this challenge are to:
Use cases focused on object detection and avoidance, object identification, and signal processing in order to verify the performance of the technologies.