Present-day industrial robots are made for the purpose of repeating several tasks thousands of times. What the manufacturing industry needs instead is a robot that can do thousands of tasks, a few times. Programming a robot to solve just one complex motor task has remained a challenging, costly and time-consuming task. In fact, it has become the key bottleneck in industrial robotics. Empowering robots with the ability to autonomously learn such tasks is a promising approach, and, in theory, machine learning has promised fully adaptive control algorithms which learn both by observation and trial-and-error. However, to date, learning techniques have yet to fulfil this promise, as only few methods manage to scale into the high-dimensional domains of manipulator robotics, or even the new upcoming trend of collaborative robots. The goal of the AssemblySkills ERC PoC is to validate an autonomous skill learning system that enables industrial robots to acquire and improve a rich set of motor skills. Using structured, modular control architectures is a promising concept to scale robot learning to more complex real-world tasks. In such a modular control architecture, elemental building blocks – called movement primitives, need to be adapted, sequenced or co-activated simultaneously. Within the ERC PoC AssemblySkills, our goal is to group these modules into an industry-scale complete software package that can enable industrial robots to learn new skills (particularly in assembly tasks). The value proposition of our ERC PoC is a cost-effective, novel machine learning system that can unlock the potential of manufacturing robots by enabling them to learn to select, adapt and sequence parametrized building blocks such as movement primitives. Our approach is unique in the sense that it can acquire more than just a single desired trajectory as done in competing approaches, capable of save policy adaptation, requires only little data and can explain the solution to the robot operator.