Federated learning for robotics and automation is a rapidly evolving research area that focuses on enabling multiple robots or automated systems to collaboratively learn models without sharing raw sensory or operational data, preserving privacy and improving scalability. This approach is particularly valuable in multi-robot systems, industrial automation, and smart manufacturing, where heterogeneous environments and limited connectivity pose challenges for centralized learning. Research explores integrating federated learning with deep reinforcement learning, convolutional neural networks (CNNs), recurrent neural networks (RNNs), and graph neural networks (GNNs) to enhance perception, control, and decision-making tasks. Key topics include handling non-i.i.d. and heterogeneous data, communication-efficient model aggregation, personalization for individual robots, and robustness against adversarial attacks or system failures. Applications include collaborative navigation, object manipulation, industrial process optimization, autonomous vehicles, and swarm robotics. Recent studies also investigate hybrid edge-cloud federated frameworks, real-time adaptation, and transfer learning, establishing federated learning as a promising paradigm for privacy-aware, scalable, and intelligent robotic and automation systems.