Kamil Bujnarowski

Cognitive robotics is a research field at the intersection of robotics and artificial intelligence (AI). It aims to provide robotic systems with cognitive mechanisms, allowing them to perceive their environment, process information, and make context-sensitive decisions (Aly et al., 2017; Cangelosi & Asada, 2022; Levesque & Lakemeyer, 2008). This field of research addresses several shortcomings of traditional robotics in automation. Traditional robots are effective at performing pre-programmed tasks in structured environments, but they cannot adapt to changes, handle uncertainty well, or generalize their knowledge to other contexts (Qadeer et al., 2024; Y. Wang, 2010). Industrial automation, while highly efficient in stable environments, is inherently fragile: deviations from expected input can lead to errors (Cangelosi & Asada, 2022; Galin & Meshcheryakov, 2019; Javaid et al., 2021). Traditional automation excels in controlled, repetitive, and highly predictable domains, such as assembly line production. However, it struggles with contexts characterized by inconsistency, variability, incomplete information, or the need for interaction with humans (Dmytriyev et al., 2024; Galin & Meshcheryakov, 2019).

This dissertation presents the design and implementation of a versatile cognitive robot cell, developed within the European Horizon 2020 project ‘Cognitive Robotics for automated and digitised Insect Farming’ (CoRoSect). CoRoSect aims to develop a framework for the deployment of network-based cognitive robotic systems, using insect farming as a use case that allows us to map their strengths and weaknesses. The unstructured environment within an insect farm offers opportunities for valuable stress tests of cognitive robotic systems and reveals limitations: in terms of accessibility and degrees of freedom for flexible manipulation in confined spaces, for refined gripping tasks with a diversity of objects and even living organisms, regarding vulnerability to noise and occlusion in visual data affecting object observation and localization, and the need for frequent decision-making. Handling live insects in an insect farm requires robots to cope well with rapidly changing conditions, which create unique problems depending on the developmental stages of different insect species. For example, occlusion by burrowing larvae, rapid growth cycles requiring an adapted approach for each phase, or the need for continuous quality inspection.

The goal of this dissertation is to gain insight into the versatility provided by cognitive robotic systems, thereby addressing the limitations of a collaborative, articulated robot arm to increase its scope and capabilities in a non-highly-controlled environment. More specifically, the contributions presented in this dissertation are aimed at addressing challenges for the automation of unstructured, human-centered workspaces in which robots handle material, perform quality management, and execute fine operations. The main results and contributions, highlighted in each chapter, are: the design and control software of an extension of the kinematic chain for a larger working range and greater freedom of action in confined spaces (Chapter 3), a soft-and-rigid hybrid end-effector for versatile material handling (Chapter 4), the structural design of the robot cell and the implementation of the control software (Chapter 5), and the implementation of cognitive capabilities in perception, object localization, and error correction strategies, along with experiments and successful demonstrations of use cases in material handling and quality management (Chapter 6). The robot has been implemented as a prototype, tested, and validated based on end-user requirements within CoRoSect. The versatility of the cognitive robot is demonstrated by a variety of unique use cases from three different insect species, namely Tenebrio Molitor (mealworms), Hermetia Illucens (black soldier flies), and Acheta Domesticus (house crickets). The results and technologies are discussed in the context of the CoRoSect framework, which proposes a robotization solution that has proven capable of handling processes from three of the most farmed insect species in the EU. These chapters are structured around the development of the key components of a cognitive robot cell.