Task and Motion Planning For Human-Robot Collaboration in a Dynamic Industrial Scenario

The transition from Industry 4.0 to Industry 5.0 marks a shift from simple process automation to seamless collaboration between humans and robots. Industry 5.0 emphasizes an industrial environment centered on human needs, where operators and robots share the same workspace while ensuring safety and trust. However, planning actions and movements in such dynamic environments remains a significant challenge, as uncertainties and disturbances influence both perception and execution. Completing complex manipulation tasks in crowded settings requires carefully integrating reasoning and motion planning, a problem typically addressed through Task and Motion Planning (TAMP). Thus, TAMP is crucial for effective Human-Robot Collaboration (HRC) in industrial applications. However, existing methods struggle to accommodate human interventions and adapt to rapidly changing environments, highlighting the need for more advanced solutions. This thesis addresses the problem of TAMP in a HRC industrial scenario, where the system has to manage the worker and the uncertainty related to his/her presence. The overall contribution of this thesis is the Human Adaptive Task and Motion Planning (HAD-TAMP) framework, which seamlessly integrates human poses and actions into the planning process to quickly adapt to human requests or deviations from the process plan. The framework is constructed around three core modules. The task planning module generates and updates task sequences in real time based on human input, ensuring seamless adaptation to dynamic conditions. The motion planning module consists of specialized planners tailored to different phases of collaboration, such as material transportation, optimizing movement efficiency and safety. Furthermore, this module investigates dynamic role allocation policies to manage the partnership between operator and robot and safety methods to ensure collision-free trajectory adjustment. Finally, the context reasoner module integrates sensory data to coordinate the entire process, enabling informed decision-making and adaptive control throughout the interaction. The proposed system is extensively tested in a real industrial HRC scenario through two applications: gesture-based human-robot interaction and close collaboration in carbon fiber draping. Experimental results confirm the framework’s effectiveness in adapting to diverse human inputs while maintaining high efficiency. Notably, the re-planning process is four to five times faster than generating a new plan, demonstrating its ability to respond swiftly to dynamic operational demands.