MAGICIAN · iMmersive leArninG for ImperfeCtion detectIon and repAir through human-robot interactioN

A common trait of many important markets is the increasing attention of consumers to the aesthetic quality of the products. Even productslike mid-segment cars are required to be defect-free in all the areas falling under the direct sight of the customer. These expectations translate into high-quality standards in the production process, which are currently met by requiring an important physical effort to the workers in unsafe environments. The MAGICIAN project will take on the challenge producing a modular automation solution in which robots are used to detect and rework production defects before the last production phases commence and the aesthetics of the product is finalised. The project will produce two robotic solutions, one for defect analysis (the SR) and one for the defects’ rework (the CR). The SR and the CR can be used separately, with the humans remaining in charge of some of the activities, or in combination, with the CR operating on the defects identified by the SR. The SR can also be used in connection with the welding robotic station in order to adapt the process parameters. The robots will use Artificial Intelligence modules to detect and discriminate the defects from multi-modal data (the SR) or to decide the best policy to use for defect rework (the CR). In both cases, the decision logic of the modules will be trained using machine learning algorithms. The training data set will be acquired with the help of workers, who will operate on semi-worked products within a controlled environment. The SR and the CR will rely on the software services of a common robotic platform. The solution will be developed adopting a human-centered approach, which will allow us to evaluate the impact of the innovation on the production processes and remove the most important asperities along this path. The effectiveness of the solution will be tested on a use-case, and its generality proven by recruiting additional contributors and use-cases through a FSTP scheme.