Planning to Interact: Shaping the Intelligence of Collaborative Robots in Interaction-Rich Shared Environments

As automation and artificial intelligence become ever more prevalent, robots gradually leave the controlled confines of factories and wander into more free-ranging fields. Collaborative robotics represents the key enabling technology driving this transformation, as the improvements in their computational and physical intelligence make deploying heterogeneous robot teams in real-world environments, possibly shared with humans, an ever-closer reality. Nevertheless, for this new reality to take hold, collaborative robots should be provided with new and improved decision-making capabilities for planning and control to cope with challenges posed by real-world environments. Classically, robots have proven to excel in executing repetitive tasks in structured and controlled environments. However, real-world environments are inherently unstructured, dynamic, and uncertain. Decisions should be taken based on partial information about the state of the world, and the outcomes of action can be difficult to predict. In addition, the environment can be shared with other agents, with whom collaboration and coordination may be necessary to accomplish tasks. This requires complex algorithms and software to enable the robots to communicate and collaborate effectively. Additionally, there are often safety concerns when using multiple robots in close proximity to each other and humans. Humans can adapt to these challenging environments and situations, while robots struggle to achieve comparable effectiveness. Endowing robots with such flexibility and adaptability in interacting with the world is the crucial challenge researchers face today. The work reported in this dissertation has been motivated and inspired by the quest for solutions to this challenge. In this context, this dissertation presents novel planning approaches to efficiently manage different types of interactions collaborative robots would experience in the real world: with other autonomous agents, with humans, and finally, with the environment. First, scalable solutions to handle interactions with other autonomous agents have been investigated at different planning, coordination, and control levels. A high-level task-planning framework for object manipulation with multi-robot systems is presented, addressing the problem of autonomously determining the sequence of actions each robot should take. Then, the problem of coordinating the actions and motions of multiple robots is considered, focusing on the coordination of large fleets of autonomous vehicles and a control framework for cooperative object transportation. Secondly, this dissertation addresses the safety aspects arising from the interactions of robots with humans. Different solutions to guarantee safety by providing robots with compliant elements within their mechanical and actuation structures or by planning their motion to reduce the effects of potential collisions are presented. Special attention is devoted to integrating these safety aspects with the performance requirements that robotic systems must meet. Finally, the last part of this dissertation is dedicated to presenting approaches that tackle the problem of the interactions between robots and the environment. It shows and validates human-inspired and data-driven solutions for different tasks: tabletop grasping of unknown objects, multi-object rearrangement in confined spaces, dexterous manipulation planning for picking and placing large objects with a dual-arm system, and finally, autonomous plastic film cutting. The developed concepts are validated using state-of-the-art robots in scenarios and tasks typical of the logistic sector and warehouse operations. Logistics represent a relevant use case for validating the proposed methods, as the sector is seeing an increasing demand for highly flexible, human-aware, and efficient autonomous solutions that have to be scalable, safe, and adaptive to environmental changes