In the context of rapidly evolving technology and increased attention to social and environmental dimensions, the industrial sector is transitioning from Industry 4.0 to Industry 5.0. This new paradigm emphasizes human centrality in highly automated environments, necessitating the exploration of collaboration mechanisms between humans and robots. This study investigates the application of Physics-Informed Neural Networks (PINNs) to enhance Human-Robot Collaboration (HRC). PINNs integrates the traditional data-driven approach based on machine learning models with a prior physical knowledge of the system, providing a valuable solution when data are scarce or physical models are too complex or incomplete. The study first focuses on the main aspects of interest in the field of HRC through a state-of-the-art analysis, evaluating the application of Physics-Informed Neural Networks (PINNs) in HRC and robotics. It then delves into the definition of PINNs and their implementation. Finally, as a proof of concept, the model is applied to a case study concerning collision detection in a 6 DoF robotic arm. This is achieved by predicting the joint currents and comparing them with the measured values to identify the contribution due to external forces such as collisions. The results demonstrate that the PINNs model outperforms a traditional neural network, achieving an average error below 10%. Additionally, the collision detection application shows an f1_score of 0.80, indicating strong performance.