AI-PackBot: Inteligência artificial para packing de objectos diversificados para robots colaborativos

As society ages, the demand for personalized products grows and e-commerce booms, we face new challenges and opportunities which expand beyond existing robotic capabilities. In many work environments, humans are still tackling hazardous, mundane, and repetitive tasks. Rather than robots assisting humans, it's usually the other way around - people aiding robots. For instance, in factories, humans are often needed to sort and package diverse items that robots can't handle due to their complexity. We also see humans on automated production lines performing tasks demanding higher dexterity. Our prime goal is to empower robots with advanced perception, manipulation, and reasoning abilities to effectively assist humans in challenging tasks. We aim to achieve this through devising innovative systems and algorithms that can interact with humans, utilizing their superior cognitive abilities and contextual awareness. These systems will predict outcomes of various object interactions and be capable of identifying and manipulating objects in cluttered scenarios. This research is centered on four critical research questions (RQ): RQ1: How do we improve the ability of robots to detect and locate objects in cluttered environments and plan to pick up, handle, and place these objects amidst others? RQ2: How can we understand, model, and predict the semantic relationships between multiple objects and anticipate the effects of manipulating them? RQ3: What is the best way for users to teach robots efficiently through interaction, comfortably, and ergonomically? What are the most effective sensory modalities and interfaces for establishing shared task consciousness? RQ4: How can we make the most out of human operators' presence during tasks, providing them contextual information to lessen their cognitive and physical load? How can we monitor task uncertainty and employ such data to avoid hazards and recognize when to seek help? Beyond its distinct scientific implications tied to research questions, PackBOT will significantly bolster health and safety practices in industrial settings. Additionally, it will streamline operations across several sectors including logistics, e-commerce, low-volume manufacturing, and packaging. We envision two distinct use-case scenarios with varying complexity levels. The first scenario: an industrial scenario, involves packaging kitchenware within the smallest possible volume. Though the task becomes relatively less complex as the object characteristics are predictable, it does present distinct challenges in terms of volume constraints and fragility. The second scenario envisions larger-scale packing projects involving a diverse array of standard household items typically found during grocery shopping. Here, efficiency in packing becomes less crucial but the complexity increases as the shape and appearance of objects are unknown. Also, additional considerations regarding thefragility or potential contamination must be taken into account when packaging combinations.