Abstract

This paper presents a proposal for integrating Graph Convolutional Neural Network (Graph-CNN), Multimodal Transformer, Soft Actor-Critic with Reward Constrained Policy Optimisation (SAC-RCPO), and TactileGAN algorithms, which are used in human-robot collaboration. This integration addresses challenges such as adaptability, safety, intention recognition, and precise manipulation in dynamic environments. The central hypothesis is that the combination of these algorithms could lead to improvements in manipulation accuracy, collision avoidance, and training efficiency. The methodology evaluates this combination in simulated and controlled laboratory environments that replicate industrial situations. This work considers the human dimension of collaboration, emphasising the importance of interaction, user trust, and ergonomic integration. It also recognises the ethical implications related to safety, transparency, and the responsible implementation of Artificial Intelligence (AI) systems in shared workspaces. The proposed framework has potential applications in the manufacturing, healthcare, and service sectors, aiming to drive the practical adoption of safe and efficient collaborative robots.