Titolo della tesi: Multimodal AI Approach to Address Complex Problems in a Real Environments
This thesis stems from some of the work I did as Research Technology Manager in the Konica Minolta company and, in the past, as a Research Fellow in the ALCOR Lab, Sapienza University of Rome, working for H2020 Second Hands project. Con- sequently, this work has an industrial nature and concerns the work performed with the team that I lead, which deals with developing solutions that, through multimodal perception, make it possible to provide Assistive Services in different work contexts. The thesis show how state-of-the-art algorithms and internally developed solutions can leave the laboratory and actually operate and solve problems in dynamic and complex realities.
This work shows how solutions conceived and developed horizontally, are then verticalized in an appropriate manner in order to guarantee different types of services, responding to the real needing from the work reality, in particular in the Healthcare and Manufacturing verticals. The thesis is divided into two main strands, which will illustrate assistive solutions in different contexts, using different platforms.
• Safety and Security: which concerns the development and use of perception technologies in order to generate safety and security services, in fact, in different work environments. In this vein the input sources are fixed cameras of different nature (frontal RGB, spherical RGB, thermal cameras, night lenses) and other sensory types, such as microphones or vibrometers.
• Assistive Robotics: which concerns the development and use of a cognitive system and perceptual solutions on robotic platforms of different nature (humanoids and cobots), used both as an input source and as a means of active or proactive reaction to situations in various contexts.
All the technologies, algorithms and solutions described in this thesis have been created following the Konica Minolta Technology Maturity Level (TML) development process and Konica Minolta Stage Gate process, which from the preliminary conception and development, passes through different stages of Definition of Done and Quality and Assurance, both on the software side and on the artifacts of Machine / Deep Learning, up to the test in the real environment and business validation. Consequently, all the solutions described are ranked at level 9 of the Technology Readiness Level (TRL) scale.