Research of autonomous manufacturing systems is motivated both by the new technical possibilities of cyber-physical systems and by the practical needs of the industry. Autonomous operation in semi-structured industrial environments can now be supported by advanced sensor technologies, digital twins, artificial intelligence and novel communication techniques. These enable real-time monitoring of production processes, situation recognition and prediction, automated and adaptive (re)planning, teamwork and performance improvement by learning. This paper summarizes the main requirements towards autonomous industrial robotics and suggests a generic workflow for realizing such systems. Application case studies will be presented from recent practice at HUN-REN SZTAKI in a broad range of domains such as assembly, welding, grinding, picking and placing, and machining. The various solutions have in common that they use a generic digital twin concept as their core. After making general recommendations for realizing autonomous robotic solutions in the industry, open issues for future research will be discussed.