ROSSINI will provide several advances beyond the state of the art, developing a Safety Aware Control Architecture for robot cognitive perception and optimal task planning and execution. In order to enable actual perception, ROSSINI will deploy a data processing technique based on artificial intelligence, taking advantage of GPU computing and advanced machine learning algorithms for real-time image recognition, to obtain a semantic scene map that adapts to dynamic working conditions. ROSSINI will adopt AI algorithms also for motion prediction of humans and moving entities, and will embed their stochastic information in the dynamic semantic scene map. This will allow the robot to further refine its planning in order to maximize performance while preserving safety. Moreover, the deployment of artificial intelligence techniques for data processing will not be just aimed to enable robot perception, but will also make robot cognition possible. ROSSINI Safety Aware Control Architecture will make it possible for robots to optimally schedule the tasks reacting to a changing environment. Each action to be executed will be sent to a dynamic planner that will dynamically optimize its execution in terms of trajectory to follow and/or interactive behaviour to reproduce while considering the variable safety conditions in the working area. In this way, ROSSINI will reach the goal to optimise the trade-off between human operator safety and manufacturing productivity and will introduce a novel perception and control architecture blending safety and performance oriented planning/control.